You need to download the acoustic model and the vocoder model.

Download the acoustic model

Please use the following code to download the model:

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/matcha-icefall-zh-en.tar.bz2

tar xvf matcha-icefall-zh-en.tar.bz2
rm matcha-icefall-zh-en.tar.bz2

You should see the following output:

ls -lh matcha-icefall-zh-en/
total 168432
-rw-r--r--@   1 fangjun  staff    58K  4 Dec 14:29 date-zh.fst
drwxr-xr-x@ 122 fangjun  staff   3.8K 28 Nov  2023 espeak-ng-data
-rw-r--r--@   1 fangjun  staff   1.3M  4 Dec 14:29 lexicon.txt
-rw-r--r--@   1 fangjun  staff    72M  4 Dec 14:29 model-steps-3.onnx
-rw-r--r--@   1 fangjun  staff    63K  4 Dec 14:29 number-zh.fst
-rw-r--r--@   1 fangjun  staff    87K  4 Dec 14:29 phone-zh.fst
-rw-r--r--@   1 fangjun  staff   2.0K  4 Dec 14:29 README.md
-rw-r--r--@   1 fangjun  staff    21K  4 Dec 14:29 tokens.txt

Download the vocoder model

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/vocos-16khz-univ.onnx

You should see the following output

ls -lh vocos-16khz-univ.onnx

-rw-r--r--@ 1 fangjun  staff    51M  4 Dec 14:54 vocos-16khz-univ.onnx

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

The following code shows how to use the Python API of sherpa-onnx with this model.

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        matcha=sherpa_onnx.OfflineTtsMatchaModelConfig(
            acoustic_model="matcha-icefall-zh-en/model-steps-3.onnx",
            vocoder="vocos-16khz-univ.onnx",
            lexicon="matcha-icefall-zh-en/lexicon.txt",
            tokens="matcha-icefall-zh-en/tokens.txt",
            data_dir="matcha-icefall-zh-en/espeak-ng-data",
        ),
        num_threads=2,
        debug=True, # set it False to disable debug output
    ),
    max_num_sentences=1,
    rule_fsts="matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst",
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。"


audio = tts.generate(text, sid=0, speed=1.0)

sf.write(
    "./test.mp3",
    audio.samples,
    samplerate=audio.sample_rate,
)

You can save it as test_zh_en.py and then run:

pip install sherpa-onnx soundfile

python3 ./test_zh_en.py

You will get a file test.mp3 in the end.

You can use the following code to play with matcha-icefall-zh-en using C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.matcha.acoustic_model = "matcha-icefall-zh-en/model-steps-3.onnx";
  config.model.matcha.vocoder = "vocos-16khz-univ.onnx";
  config.model.matcha.lexicon = "matcha-icefall-zh-en/lexicon.txt";
  config.model.matcha.tokens = "matcha-icefall-zh-en/tokens.txt";
  config.model.matcha.data_dir = "matcha-icefall-zh-en/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;
  config.rule_fsts = "matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-zh-en.c. Then you can compile it with the following command:

gcc \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-zh-en \
  /tmp/test-zh-en.c

Now you can run

cd /tmp

# Assume you have downloaded the acoustic model as well as the vocoder model and put them to /tmp
./test-zh-en

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-zh-en.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with matcha-icefall-zh-en using C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.matcha.acoustic_model = "matcha-icefall-zh-en/model-steps-3.onnx";
  config.model.matcha.vocoder = "vocos-16khz-univ.onnx";
  config.model.matcha.lexicon = "matcha-icefall-zh-en/lexicon.txt";
  config.model.matcha.tokens = "matcha-icefall-zh-en/tokens.txt";
  config.model.matcha.data_dir = "matcha-icefall-zh-en/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;
  config.rule_fsts = "matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst";

  std::string filename = "./test.wav";
  std::string text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-zh-en.cc. Then you can compile it with the following command:

g++ \
  -std=c++17 \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-cxx-api \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-zh-en \
  /tmp/test-zh-en.cc

Now you can run

cd /tmp

# Assume you have downloaded the acoustic model as well as the vocoder model and put them to /tmp
./test-zh-en

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-zh-en.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with matcha-icefall-zh-en with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsMatchaModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            matcha: OfflineTtsMatchaModelConfig {
                acoustic_model: Some("matcha-icefall-zh-en/model-steps-3.onnx".into()),
                vocoder: Some("vocos-16khz-univ.onnx".into()),
                tokens: Some("matcha-icefall-zh-en/tokens.txt".into()),
                data_dir: Some("matcha-icefall-zh-en/espeak-ng-data".into()),
                lexicon: Some("matcha-icefall-zh-en/lexicon.txt".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        rule_fsts: Some("matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst".into()),
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with matcha-icefall-zh-en with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      matcha: {
        acousticModel: 'matcha-icefall-zh-en/model-steps-3.onnx',
        vocoder: 'vocos-16khz-univ.onnx',
        tokens: 'matcha-icefall-zh-en/tokens.txt',
        dataDir: 'matcha-icefall-zh-en/espeak-ng-data',
        lexicon: 'matcha-icefall-zh-en/lexicon.txt',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
    ruleFsts: 'matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst',
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = '我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final matcha = sherpa_onnx.OfflineTtsMatchaModelConfig(
    acousticModel: 'matcha-icefall-zh-en/model-steps-3.onnx',
    vocoder: 'vocos-16khz-univ.onnx',
    tokens: 'matcha-icefall-zh-en/tokens.txt',
    dataDir: 'matcha-icefall-zh-en/espeak-ng-data',
    lexicon: 'matcha-icefall-zh-en/lexicon.txt',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    matcha: matcha,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: '我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Swift API.

func run() {
  let matcha = sherpaOnnxOfflineTtsMatchaModelConfig(
    acousticModel: "matcha-icefall-zh-en/model-steps-3.onnx",
    vocoder: "vocos-16khz-univ.onnx",
    tokens: "matcha-icefall-zh-en/tokens.txt",
    dataDir: "matcha-icefall-zh-en/espeak-ng-data",
    lexicon: "matcha-icefall-zh-en/lexicon.txt"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(matcha: matcha)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Matcha.AcousticModel = "matcha-icefall-zh-en/model-steps-3.onnx";
config.Model.Matcha.Vocoder = "vocos-16khz-univ.onnx";
config.Model.Matcha.Tokens = "matcha-icefall-zh-en/tokens.txt";
config.Model.Matcha.DataDir = "matcha-icefall-zh-en/espeak-ng-data";
config.Model.Matcha.Lexicon = "matcha-icefall-zh-en/lexicon.txt";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.RuleFsts = "matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      matcha = OfflineTtsMatchaModelConfig(
        acousticModel = "matcha-icefall-zh-en/model-steps-3.onnx",
        vocoder = "vocos-16khz-univ.onnx",
        tokens = "matcha-icefall-zh-en/tokens.txt",
        dataDir = "matcha-icefall-zh-en/espeak-ng-data",
        lexicon = "matcha-icefall-zh-en/lexicon.txt",
      ),
      numThreads = 1,
      debug = true,
    ),
    ruleFsts = "matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst",
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var matcha = new OfflineTtsMatchaModelConfig();
    matcha.setAcousticModel("matcha-icefall-zh-en/model-steps-3.onnx");
    matcha.setVocoder("vocos-16khz-univ.onnx");
    matcha.setTokens("matcha-icefall-zh-en/tokens.txt");
    matcha.setDataDir("matcha-icefall-zh-en/espeak-ng-data");
    matcha.setLexicon("matcha-icefall-zh-en/lexicon.txt");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setMatcha(matcha);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setRuleFsts("matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst");
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Pascal API.

program test_matcha;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Matcha.AcousticModel := 'matcha-icefall-zh-en/model-steps-3.onnx';
  Config.Model.Matcha.Vocoder := 'vocos-16khz-univ.onnx';
  Config.Model.Matcha.Tokens := 'matcha-icefall-zh-en/tokens.txt';
  Config.Model.Matcha.DataDir := 'matcha-icefall-zh-en/espeak-ng-data';
  Config.Model.Matcha.Lexicon := 'matcha-icefall-zh-en/lexicon.txt';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.RuleFsts := 'matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst';
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with matcha-icefall-zh-en with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Matcha: sherpa.OfflineTtsMatchaModelConfig{
				AcousticModel: "matcha-icefall-zh-en/model-steps-3.onnx",
				Vocoder:       "vocos-16khz-univ.onnx",
				Tokens:        "matcha-icefall-zh-en/tokens.txt",
				DataDir:       "matcha-icefall-zh-en/espeak-ng-data",
				Lexicon:       "matcha-icefall-zh-en/lexicon.txt",
			},
			NumThreads: 1,
			Debug:      true,
		},
		RuleFsts: "matcha-icefall-zh-en/phone-zh.fst,matcha-icefall-zh-en/date-zh.fst,matcha-icefall-zh-en/number-zh.fst",
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。在这次vocation中，我们计划去Paris欣赏埃菲尔铁塔和卢浮宫的美景。某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。开始数字测试。2025年12月4号，拨打110或者189202512043。123456块钱。在这个快速发展的时代，人工智能技术正在改变我们的生活方式。语音合成作为人工智能的重要应用之一，让机器能够用自然流畅的语音与人类进行交流。"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-multi-lang-v1_0.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-multi-lang-v1_0.tar.bz2

You can use the following code to play with kokoro-multi-lang-v1_0

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        kokoro=sherpa_onnx.OfflineTtsKokoroModelConfig(
            model="kokoro-multi-lang-v1_0/model.onnx",
            voices="kokoro-multi-lang-v1_0/voices.bin",
            tokens="kokoro-multi-lang-v1_0/tokens.txt",
            data_dir="kokoro-multi-lang-v1_0/espeak-ng-data",
            lexicon="kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with kokoro-multi-lang-v1_0 with C API.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));
  config.model.kokoro.model = "kokoro-multi-lang-v1_0/model.onnx";
  config.model.kokoro.voices = "kokoro-multi-lang-v1_0/voices.bin";
  config.model.kokoro.tokens = "kokoro-multi-lang-v1_0/tokens.txt";
  config.model.kokoro.data_dir = "kokoro-multi-lang-v1_0/espeak-ng-data";
  config.model.kokoro.lexicon = "kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt";

  config.model.num_threads = 1;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 0;

  const char *text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

#if 0
  // If you don't want to use a callback, then please enable this branch
  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);
#else
  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg,
                                             ProgressCallback, NULL);
#endif

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-kokoro.c. Then you can compile it with the following command:

gcc \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-kokoro \
  /tmp/test-kokoro.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-kokoro

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-kokoro.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with kokoro-multi-lang-v1_0 with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;

  config.model.kokoro.model = "kokoro-multi-lang-v1_0/model.onnx";
  config.model.kokoro.voices = "kokoro-multi-lang-v1_0/voices.bin";
  config.model.kokoro.tokens = "kokoro-multi-lang-v1_0/tokens.txt";
  config.model.kokoro.data_dir = "kokoro-multi-lang-v1_0/espeak-ng-data";
  config.model.kokoro.lexicon = "kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt";

  config.model.num_threads = 1;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-kokoro.cc. Then you can compile it with the following command:

g++ \
  -std=c++17 \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-cxx-api \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-kokoro \
  /tmp/test-kokoro.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-kokoro

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-kokoro.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with kokoro-multi-lang-v1_0 with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsKokoroModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            kokoro: OfflineTtsKokoroModelConfig {
                model: Some("kokoro-multi-lang-v1_0/model.onnx".into()),
                voices: Some("kokoro-multi-lang-v1_0/voices.bin".into()),
                tokens: Some("kokoro-multi-lang-v1_0/tokens.txt".into()),
                data_dir: Some("kokoro-multi-lang-v1_0/espeak-ng-data".into()),
                lexicon: Some("kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with kokoro-multi-lang-v1_0 with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      kokoro: {
        model: 'kokoro-multi-lang-v1_0/model.onnx',
        voices: 'kokoro-multi-lang-v1_0/voices.bin',
        tokens: 'kokoro-multi-lang-v1_0/tokens.txt',
        dataDir: 'kokoro-multi-lang-v1_0/espeak-ng-data',
        lexicon: 'kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final kokoro = sherpa_onnx.OfflineTtsKokoroModelConfig(
    model: 'kokoro-multi-lang-v1_0/model.onnx',
    voices: 'kokoro-multi-lang-v1_0/voices.bin',
    tokens: 'kokoro-multi-lang-v1_0/tokens.txt',
    dataDir: 'kokoro-multi-lang-v1_0/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    kokoro: kokoro,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Swift API.

func run() {
  let kokoro = sherpaOnnxOfflineTtsKokoroModelConfig(
    model: "kokoro-multi-lang-v1_0/model.onnx",
    voices: "kokoro-multi-lang-v1_0/voices.bin",
    tokens: "kokoro-multi-lang-v1_0/tokens.txt",
    dataDir: "kokoro-multi-lang-v1_0/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(kokoro: kokoro)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Kokoro.Model = "kokoro-multi-lang-v1_0/model.onnx";
config.Model.Kokoro.Voices = "kokoro-multi-lang-v1_0/voices.bin";
config.Model.Kokoro.Tokens = "kokoro-multi-lang-v1_0/tokens.txt";
config.Model.Kokoro.DataDir = "kokoro-multi-lang-v1_0/espeak-ng-data";
config.Model.Kokoro.Lexicon = "kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = ;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      kokoro = OfflineTtsKokoroModelConfig(
        model = "kokoro-multi-lang-v1_0/model.onnx",
        voices = "kokoro-multi-lang-v1_0/voices.bin",
        tokens = "kokoro-multi-lang-v1_0/tokens.txt",
        dataDir = "kokoro-multi-lang-v1_0/espeak-ng-data",
        lexicon = "kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = ,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var kokoro = new OfflineTtsKokoroModelConfig();
    kokoro.setModel("kokoro-multi-lang-v1_0/model.onnx");
    kokoro.setVoices("kokoro-multi-lang-v1_0/voices.bin");
    kokoro.setTokens("kokoro-multi-lang-v1_0/tokens.txt");
    kokoro.setDataDir("kokoro-multi-lang-v1_0/espeak-ng-data");
    kokoro.setLexicon("kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setKokoro(kokoro);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Pascal API.

program test_kokoro;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Kokoro.Model := 'kokoro-multi-lang-v1_0/model.onnx';
  Config.Model.Kokoro.Voices := 'kokoro-multi-lang-v1_0/voices.bin';
  Config.Model.Kokoro.Tokens := 'kokoro-multi-lang-v1_0/tokens.txt';
  Config.Model.Kokoro.DataDir := 'kokoro-multi-lang-v1_0/espeak-ng-data';
  Config.Model.Kokoro.Lexicon := 'kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_0 with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Kokoro: sherpa.OfflineTtsKokoroModelConfig{
				Model:  "kokoro-multi-lang-v1_0/model.onnx",
				Voices: "kokoro-multi-lang-v1_0/voices.bin",
				Tokens: "kokoro-multi-lang-v1_0/tokens.txt",
				DataDir: "kokoro-multi-lang-v1_0/espeak-ng-data",
				Lexicon: "kokoro-multi-lang-v1_0/lexicon-us-en.txt,kokoro-multi-lang-v1_0/lexicon-zh.txt",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号."

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-multi-lang-v1_1.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-multi-lang-v1_1.tar.bz2

You can use the following code to play with kokoro-multi-lang-v1_1

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        kokoro=sherpa_onnx.OfflineTtsKokoroModelConfig(
            model="kokoro-multi-lang-v1_1/model.onnx",
            voices="kokoro-multi-lang-v1_1/voices.bin",
            tokens="kokoro-multi-lang-v1_1/tokens.txt",
            data_dir="kokoro-multi-lang-v1_1/espeak-ng-data",
            lexicon="kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with kokoro-multi-lang-v1_1 with C API.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));
  config.model.kokoro.model = "kokoro-multi-lang-v1_1/model.onnx";
  config.model.kokoro.voices = "kokoro-multi-lang-v1_1/voices.bin";
  config.model.kokoro.tokens = "kokoro-multi-lang-v1_1/tokens.txt";
  config.model.kokoro.data_dir = "kokoro-multi-lang-v1_1/espeak-ng-data";
  config.model.kokoro.lexicon = "kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt";

  config.model.num_threads = 1;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 0;

  const char *text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

#if 0
  // If you don't want to use a callback, then please enable this branch
  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);
#else
  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg,
                                             ProgressCallback, NULL);
#endif

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-kokoro.c. Then you can compile it with the following command:

gcc \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-kokoro \
  /tmp/test-kokoro.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-kokoro

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-kokoro.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with kokoro-multi-lang-v1_1 with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;

  config.model.kokoro.model = "kokoro-multi-lang-v1_1/model.onnx";
  config.model.kokoro.voices = "kokoro-multi-lang-v1_1/voices.bin";
  config.model.kokoro.tokens = "kokoro-multi-lang-v1_1/tokens.txt";
  config.model.kokoro.data_dir = "kokoro-multi-lang-v1_1/espeak-ng-data";
  config.model.kokoro.lexicon = "kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt";

  config.model.num_threads = 1;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-kokoro.cc. Then you can compile it with the following command:

g++ \
  -std=c++17 \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-cxx-api \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-kokoro \
  /tmp/test-kokoro.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-kokoro

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-kokoro.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with kokoro-multi-lang-v1_1 with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsKokoroModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            kokoro: OfflineTtsKokoroModelConfig {
                model: Some("kokoro-multi-lang-v1_1/model.onnx".into()),
                voices: Some("kokoro-multi-lang-v1_1/voices.bin".into()),
                tokens: Some("kokoro-multi-lang-v1_1/tokens.txt".into()),
                data_dir: Some("kokoro-multi-lang-v1_1/espeak-ng-data".into()),
                lexicon: Some("kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with kokoro-multi-lang-v1_1 with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      kokoro: {
        model: 'kokoro-multi-lang-v1_1/model.onnx',
        voices: 'kokoro-multi-lang-v1_1/voices.bin',
        tokens: 'kokoro-multi-lang-v1_1/tokens.txt',
        dataDir: 'kokoro-multi-lang-v1_1/espeak-ng-data',
        lexicon: 'kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final kokoro = sherpa_onnx.OfflineTtsKokoroModelConfig(
    model: 'kokoro-multi-lang-v1_1/model.onnx',
    voices: 'kokoro-multi-lang-v1_1/voices.bin',
    tokens: 'kokoro-multi-lang-v1_1/tokens.txt',
    dataDir: 'kokoro-multi-lang-v1_1/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    kokoro: kokoro,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Swift API.

func run() {
  let kokoro = sherpaOnnxOfflineTtsKokoroModelConfig(
    model: "kokoro-multi-lang-v1_1/model.onnx",
    voices: "kokoro-multi-lang-v1_1/voices.bin",
    tokens: "kokoro-multi-lang-v1_1/tokens.txt",
    dataDir: "kokoro-multi-lang-v1_1/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(kokoro: kokoro)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Kokoro.Model = "kokoro-multi-lang-v1_1/model.onnx";
config.Model.Kokoro.Voices = "kokoro-multi-lang-v1_1/voices.bin";
config.Model.Kokoro.Tokens = "kokoro-multi-lang-v1_1/tokens.txt";
config.Model.Kokoro.DataDir = "kokoro-multi-lang-v1_1/espeak-ng-data";
config.Model.Kokoro.Lexicon = "kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = ;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      kokoro = OfflineTtsKokoroModelConfig(
        model = "kokoro-multi-lang-v1_1/model.onnx",
        voices = "kokoro-multi-lang-v1_1/voices.bin",
        tokens = "kokoro-multi-lang-v1_1/tokens.txt",
        dataDir = "kokoro-multi-lang-v1_1/espeak-ng-data",
        lexicon = "kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = ,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var kokoro = new OfflineTtsKokoroModelConfig();
    kokoro.setModel("kokoro-multi-lang-v1_1/model.onnx");
    kokoro.setVoices("kokoro-multi-lang-v1_1/voices.bin");
    kokoro.setTokens("kokoro-multi-lang-v1_1/tokens.txt");
    kokoro.setDataDir("kokoro-multi-lang-v1_1/espeak-ng-data");
    kokoro.setLexicon("kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setKokoro(kokoro);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Pascal API.

program test_kokoro;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Kokoro.Model := 'kokoro-multi-lang-v1_1/model.onnx';
  Config.Model.Kokoro.Voices := 'kokoro-multi-lang-v1_1/voices.bin';
  Config.Model.Kokoro.Tokens := 'kokoro-multi-lang-v1_1/tokens.txt';
  Config.Model.Kokoro.DataDir := 'kokoro-multi-lang-v1_1/espeak-ng-data';
  Config.Model.Kokoro.Lexicon := 'kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号.', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with kokoro-multi-lang-v1_1 with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Kokoro: sherpa.OfflineTtsKokoroModelConfig{
				Model:  "kokoro-multi-lang-v1_1/model.onnx",
				Voices: "kokoro-multi-lang-v1_1/voices.bin",
				Tokens: "kokoro-multi-lang-v1_1/tokens.txt",
				DataDir: "kokoro-multi-lang-v1_1/espeak-ng-data",
				Lexicon: "kokoro-multi-lang-v1_1/lexicon-us-en.txt,kokoro-multi-lang-v1_1/lexicon-zh.txt",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "This model supports both Chinese and English. 小米的核心价值观是什么？答案是真诚热爱！有困难，请拨打110 或者18601200909。I am learning 机器学习. 我在研究 machine learning。What do you think 中英文说的如何呢?今天是 2025年6月18号."

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_dii-high.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_dii-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_dii-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "كيف حالك اليوم؟";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_dii-high.tar.bz2

You can use the following code to play with vits-piper-ar_JO-SA_dii-high

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx",
            data_dir="vits-piper-ar_JO-SA_dii-high/espeak-ng-data",
            tokens="vits-piper-ar_JO-SA_dii-high/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="كيف حالك اليوم؟",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_dii-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_dii-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "كيف حالك اليوم؟";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx".into()),
                tokens: Some("vits-piper-ar_JO-SA_dii-high/tokens.txt".into()),
                data_dir: Some("vits-piper-ar_JO-SA_dii-high/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "كيف حالك اليوم؟";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx',
        tokens: 'vits-piper-ar_JO-SA_dii-high/tokens.txt',
        dataDir: 'vits-piper-ar_JO-SA_dii-high/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'كيف حالك اليوم؟';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx',
    tokens: 'vits-piper-ar_JO-SA_dii-high/tokens.txt',
    dataDir: 'vits-piper-ar_JO-SA_dii-high/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'كيف حالك اليوم؟', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx",
    lexicon: "",
    tokens: "vits-piper-ar_JO-SA_dii-high/tokens.txt",
    dataDir: "vits-piper-ar_JO-SA_dii-high/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "كيف حالك اليوم؟"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx";
config.Model.Vits.Tokens = "vits-piper-ar_JO-SA_dii-high/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ar_JO-SA_dii-high/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "كيف حالك اليوم؟";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx",
        tokens = "vits-piper-ar_JO-SA_dii-high/tokens.txt",
        dataDir = "vits-piper-ar_JO-SA_dii-high/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "كيف حالك اليوم؟",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx");
    vits.setTokens("vits-piper-ar_JO-SA_dii-high/tokens.txt");
    vits.setDataDir("vits-piper-ar_JO-SA_dii-high/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "كيف حالك اليوم؟";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ar_JO-SA_dii-high/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ar_JO-SA_dii-high/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('كيف حالك اليوم؟', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_dii-high with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ar_JO-SA_dii-high/ar_JO-SA_dii-high.onnx",
				Tokens:  "vits-piper-ar_JO-SA_dii-high/tokens.txt",
				DataDir: "vits-piper-ar_JO-SA_dii-high/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "كيف حالك اليوم؟"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_miro-high.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_miro-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_miro-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "كيف حالك اليوم؟";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_miro-high.tar.bz2

You can use the following code to play with vits-piper-ar_JO-SA_miro-high

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx",
            data_dir="vits-piper-ar_JO-SA_miro-high/espeak-ng-data",
            tokens="vits-piper-ar_JO-SA_miro-high/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="كيف حالك اليوم؟",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_miro-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_miro-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "كيف حالك اليوم؟";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx".into()),
                tokens: Some("vits-piper-ar_JO-SA_miro-high/tokens.txt".into()),
                data_dir: Some("vits-piper-ar_JO-SA_miro-high/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "كيف حالك اليوم؟";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx',
        tokens: 'vits-piper-ar_JO-SA_miro-high/tokens.txt',
        dataDir: 'vits-piper-ar_JO-SA_miro-high/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'كيف حالك اليوم؟';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx',
    tokens: 'vits-piper-ar_JO-SA_miro-high/tokens.txt',
    dataDir: 'vits-piper-ar_JO-SA_miro-high/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'كيف حالك اليوم؟', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx",
    lexicon: "",
    tokens: "vits-piper-ar_JO-SA_miro-high/tokens.txt",
    dataDir: "vits-piper-ar_JO-SA_miro-high/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "كيف حالك اليوم؟"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx";
config.Model.Vits.Tokens = "vits-piper-ar_JO-SA_miro-high/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ar_JO-SA_miro-high/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "كيف حالك اليوم؟";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx",
        tokens = "vits-piper-ar_JO-SA_miro-high/tokens.txt",
        dataDir = "vits-piper-ar_JO-SA_miro-high/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "كيف حالك اليوم؟",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx");
    vits.setTokens("vits-piper-ar_JO-SA_miro-high/tokens.txt");
    vits.setDataDir("vits-piper-ar_JO-SA_miro-high/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "كيف حالك اليوم؟";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ar_JO-SA_miro-high/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ar_JO-SA_miro-high/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('كيف حالك اليوم؟', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro-high with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ar_JO-SA_miro-high/ar_JO-SA_miro-high.onnx",
				Tokens:  "vits-piper-ar_JO-SA_miro-high/tokens.txt",
				DataDir: "vits-piper-ar_JO-SA_miro-high/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "كيف حالك اليوم؟"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_miro_V2-high.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "كيف حالك اليوم؟";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-SA_miro_V2-high.tar.bz2

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx",
            data_dir="vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data",
            tokens="vits-piper-ar_JO-SA_miro_V2-high/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="كيف حالك اليوم؟",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "كيف حالك اليوم؟";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx".into()),
                tokens: Some("vits-piper-ar_JO-SA_miro_V2-high/tokens.txt".into()),
                data_dir: Some("vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "كيف حالك اليوم؟";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx',
        tokens: 'vits-piper-ar_JO-SA_miro_V2-high/tokens.txt',
        dataDir: 'vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'كيف حالك اليوم؟';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx',
    tokens: 'vits-piper-ar_JO-SA_miro_V2-high/tokens.txt',
    dataDir: 'vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'كيف حالك اليوم؟', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx",
    lexicon: "",
    tokens: "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt",
    dataDir: "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "كيف حالك اليوم؟"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx";
config.Model.Vits.Tokens = "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "كيف حالك اليوم؟";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx",
        tokens = "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt",
        dataDir = "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "كيف حالك اليوم؟",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx");
    vits.setTokens("vits-piper-ar_JO-SA_miro_V2-high/tokens.txt");
    vits.setDataDir("vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "كيف حالك اليوم؟";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ar_JO-SA_miro_V2-high/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('كيف حالك اليوم؟', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-SA_miro_V2-high with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ar_JO-SA_miro_V2-high/ar_JO-SA_miro_V2-high.onnx",
				Tokens:  "vits-piper-ar_JO-SA_miro_V2-high/tokens.txt",
				DataDir: "vits-piper-ar_JO-SA_miro_V2-high/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "كيف حالك اليوم؟"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-kareem-low.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ar_JO-kareem-low with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-kareem-low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-kareem-low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "كيف حالك اليوم؟";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-kareem-low.tar.bz2

You can use the following code to play with vits-piper-ar_JO-kareem-low

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx",
            data_dir="vits-piper-ar_JO-kareem-low/espeak-ng-data",
            tokens="vits-piper-ar_JO-kareem-low/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="كيف حالك اليوم؟",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ar_JO-kareem-low with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-kareem-low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-kareem-low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "كيف حالك اليوم؟";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ar_JO-kareem-low with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx".into()),
                tokens: Some("vits-piper-ar_JO-kareem-low/tokens.txt".into()),
                data_dir: Some("vits-piper-ar_JO-kareem-low/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "كيف حالك اليوم؟";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ar_JO-kareem-low with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx',
        tokens: 'vits-piper-ar_JO-kareem-low/tokens.txt',
        dataDir: 'vits-piper-ar_JO-kareem-low/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'كيف حالك اليوم؟';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx',
    tokens: 'vits-piper-ar_JO-kareem-low/tokens.txt',
    dataDir: 'vits-piper-ar_JO-kareem-low/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'كيف حالك اليوم؟', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx",
    lexicon: "",
    tokens: "vits-piper-ar_JO-kareem-low/tokens.txt",
    dataDir: "vits-piper-ar_JO-kareem-low/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "كيف حالك اليوم؟"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx";
config.Model.Vits.Tokens = "vits-piper-ar_JO-kareem-low/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ar_JO-kareem-low/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "كيف حالك اليوم؟";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx",
        tokens = "vits-piper-ar_JO-kareem-low/tokens.txt",
        dataDir = "vits-piper-ar_JO-kareem-low/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "كيف حالك اليوم؟",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx");
    vits.setTokens("vits-piper-ar_JO-kareem-low/tokens.txt");
    vits.setDataDir("vits-piper-ar_JO-kareem-low/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "كيف حالك اليوم؟";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ar_JO-kareem-low/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ar_JO-kareem-low/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('كيف حالك اليوم؟', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-low with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ar_JO-kareem-low/ar_JO-kareem-low.onnx",
				Tokens:  "vits-piper-ar_JO-kareem-low/tokens.txt",
				DataDir: "vits-piper-ar_JO-kareem-low/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "كيف حالك اليوم؟"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-kareem-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ar_JO-kareem-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-kareem-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-kareem-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "كيف حالك اليوم؟";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ar_JO-kareem-medium.tar.bz2

You can use the following code to play with vits-piper-ar_JO-kareem-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx",
            data_dir="vits-piper-ar_JO-kareem-medium/espeak-ng-data",
            tokens="vits-piper-ar_JO-kareem-medium/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="كيف حالك اليوم؟",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ar_JO-kareem-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx";
  config.model.vits.tokens = "vits-piper-ar_JO-kareem-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ar_JO-kareem-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "كيف حالك اليوم؟";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx".into()),
                tokens: Some("vits-piper-ar_JO-kareem-medium/tokens.txt".into()),
                data_dir: Some("vits-piper-ar_JO-kareem-medium/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "كيف حالك اليوم؟";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ar_JO-kareem-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx',
        tokens: 'vits-piper-ar_JO-kareem-medium/tokens.txt',
        dataDir: 'vits-piper-ar_JO-kareem-medium/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'كيف حالك اليوم؟';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx',
    tokens: 'vits-piper-ar_JO-kareem-medium/tokens.txt',
    dataDir: 'vits-piper-ar_JO-kareem-medium/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'كيف حالك اليوم؟', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-ar_JO-kareem-medium/tokens.txt",
    dataDir: "vits-piper-ar_JO-kareem-medium/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "كيف حالك اليوم؟"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-ar_JO-kareem-medium/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ar_JO-kareem-medium/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "كيف حالك اليوم؟";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx",
        tokens = "vits-piper-ar_JO-kareem-medium/tokens.txt",
        dataDir = "vits-piper-ar_JO-kareem-medium/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "كيف حالك اليوم؟",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx");
    vits.setTokens("vits-piper-ar_JO-kareem-medium/tokens.txt");
    vits.setDataDir("vits-piper-ar_JO-kareem-medium/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "كيف حالك اليوم؟";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ar_JO-kareem-medium/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ar_JO-kareem-medium/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('كيف حالك اليوم؟', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ar_JO-kareem-medium with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ar_JO-kareem-medium/ar_JO-kareem-medium.onnx",
				Tokens:  "vits-piper-ar_JO-kareem-medium/tokens.txt",
				DataDir: "vits-piper-ar_JO-kareem-medium/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "كيف حالك اليوم؟"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-supertonic-3-tts-int8-2026-05-11.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-supertonic-3-tts-int8-2026-05-11.tar.bz2

You can use the following code to play with supertonic-3

import sherpa_onnx
import soundfile as sf

tts_config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        supertonic=sherpa_onnx.OfflineTtsSupertonicModelConfig(
            duration_predictor="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
            text_encoder="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
            vector_estimator="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
            vocoder="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
            tts_json="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
            unicode_indexer="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
            voice_style="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
        ),
        debug=False,
        num_threads=2,
        provider="cpu",
    ),
)

tts = sherpa_onnx.OfflineTts(tts_config)

gen_config = sherpa_onnx.GenerationConfig()
gen_config.sid = 0
gen_config.num_steps = 8
gen_config.speed = 1.0
gen_config.extra["lang"] = "ar"

audio = tts.generate("هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي", gen_config)

sf.write("test.wav", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with supertonic-3 with C API.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));
  config.model.supertonic.duration_predictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
  config.model.supertonic.text_encoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
  config.model.supertonic.vector_estimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
  config.model.supertonic.vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
  config.model.supertonic.tts_json = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
  config.model.supertonic.unicode_indexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
  config.model.supertonic.voice_style = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

  config.model.num_threads = 2;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 1;

  const char *text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.num_steps = 8;
  gen_cfg.speed = 1.0;
  gen_cfg.extra = "{\"lang\": \"ar\"}";

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg,
                                             ProgressCallback, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-supertonic.c. Then you can compile it with the following command:

gcc \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-supertonic \
  /tmp/test-supertonic.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-supertonic

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-supertonic.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with supertonic-3 with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;

  config.model.supertonic.duration_predictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
  config.model.supertonic.text_encoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
  config.model.supertonic.vector_estimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
  config.model.supertonic.vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
  config.model.supertonic.tts_json = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
  config.model.supertonic.unicode_indexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
  config.model.supertonic.voice_style = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

  config.model.num_threads = 2;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 1;

  std::string filename = "./test.wav";
  std::string text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.num_steps = 8;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed
  gen_cfg.extra = R"({"lang": "ar"})";

  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-supertonic.cc. Then you can compile it with the following command:

g++ \
  -std=c++17 \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-cxx-api \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-supertonic \
  /tmp/test-supertonic.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-supertonic

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-supertonic.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with supertonic-3 with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsSupertonicModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            supertonic: OfflineTtsSupertonicModelConfig {
                duration_predictor: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx".into()),
                text_encoder: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx".into()),
                vector_estimator: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx".into()),
                vocoder: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx".into()),
                tts_json: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json".into()),
                unicode_indexer: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin".into()),
                voice_style: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي";

    let gen_config = GenerationConfig {
        sid: 0,
        num_steps: 8,
        speed: 1.0,
        extra: Some(r#"{"lang": "ar"}"#.into()),
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with supertonic-3 with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      supertonic: {
        durationPredictor: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx',
        textEncoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx',
        vectorEstimator: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx',
        vocoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx',
        ttsJson: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json',
        unicodeIndexer: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin',
        voiceStyle: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin',
      },
      debug: true,
      numThreads: 2,
      provider: 'cpu',
    },
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  numSteps: 8,
  speed: 1.0,
  extra: {lang: 'ar'},
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with supertonic-3 with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final supertonic = sherpa_onnx.OfflineTtsSupertonicModelConfig(
    durationPredictor: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx',
    textEncoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx',
    vectorEstimator: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx',
    vocoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx',
    ttsJson: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json',
    unicodeIndexer: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin',
    voiceStyle: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    supertonic: supertonic,
    numThreads: 2,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    numSteps: 8,
    speed: 1.0,
    extra: {'lang': 'ar'},
  );
  final audio = tts.generateWithConfig(text: 'هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with supertonic-3 with Swift API.

func run() {
  let supertonic = sherpaOnnxOfflineTtsSupertonicModelConfig(
    durationPredictor: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
    textEncoder: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
    vectorEstimator: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
    vocoder: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
    ttsJson: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
    unicodeIndexer: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
    voiceStyle: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(supertonic: supertonic)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.numSteps = 8
  genConfig.speed = 1.0
  genConfig.extra = ["lang": "ar"]

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with supertonic-3 with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Supertonic.DurationPredictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
config.Model.Supertonic.TextEncoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
config.Model.Supertonic.VectorEstimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
config.Model.Supertonic.Vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
config.Model.Supertonic.TtsJson = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
config.Model.Supertonic.UnicodeIndexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
config.Model.Supertonic.VoiceStyle = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

config.Model.NumThreads = 2;
config.Model.Debug = 1;
config.Model.Provider = "cpu";

var tts = new OfflineTts(config);
var text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.NumSteps = 8;
genConfig.Speed = 1.0f;
genConfig.Extra = "{\"lang\": \"ar\"}";

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with supertonic-3 with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      supertonic = OfflineTtsSupertonicModelConfig(
        durationPredictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
        textEncoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
        vectorEstimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
        vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
        ttsJson = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
        unicodeIndexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
        voiceStyle = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
      ),
      numThreads = 2,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    numSteps = 8,
    speed = 1.0f,
    extra = mapOf("lang" to "ar"),
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with supertonic-3 with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var supertonic = new OfflineTtsSupertonicModelConfig();
    supertonic.setDurationPredictor("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx");
    supertonic.setTextEncoder("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx");
    supertonic.setVectorEstimator("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx");
    supertonic.setVocoder("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx");
    supertonic.setTtsJson("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json");
    supertonic.setUnicodeIndexer("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin");
    supertonic.setVoiceStyle("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setSupertonic(supertonic);
    modelConfig.setNumThreads(2);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);

    var tts = new OfflineTts(config);
    var text = "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setNumSteps(8);
    genConfig.setSpeed(1.0f);
    genConfig.setExtra("{\"lang\": \"ar\"}");

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with supertonic-3 with Pascal API.

program test_supertonic;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Supertonic.DurationPredictor := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx';
  Config.Model.Supertonic.TextEncoder := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx';
  Config.Model.Supertonic.VectorEstimator := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx';
  Config.Model.Supertonic.Vocoder := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx';
  Config.Model.Supertonic.TtsJson := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json';
  Config.Model.Supertonic.UnicodeIndexer := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin';
  Config.Model.Supertonic.VoiceStyle := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin';

  Config.Model.NumThreads := 2;
  Config.Model.Debug := True;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.NumSteps := 8;
  GenConfig.Speed := 1.0;
  GenConfig.Extra := '{"lang": "ar"}';

  Audio := Tts.GenerateWithConfig('هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with supertonic-3 with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Supertonic: sherpa.OfflineTtsSupertonicModelConfig{
				DurationPredictor: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
				TextEncoder:       "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
				VectorEstimator:   "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
				Vocoder:           "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
				TtsJson:           "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
				UnicodeIndexer:    "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
				VoiceStyle:        "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
			},
			NumThreads: 2,
			Debug:      true,
		},
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "هذا هو محرك تحويل النص إلى كلام باستخدام الجيل القادم من كالدي"

	genConfig := sherpa.GenerationConfig{
		Sid:       0,
		NumSteps:  8,
		Speed:     1.0,
		Extra:     `{"lang": "ar"}`,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-sq_AL-edon-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-sq_AL-edon-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx";
  config.model.vits.tokens = "vits-piper-sq_AL-edon-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-sq_AL-edon-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Çdo fillim është i vështirë, por çdo fund është i bukur.";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-sq_AL-edon-medium.tar.bz2

You can use the following code to play with vits-piper-sq_AL-edon-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx",
            data_dir="vits-piper-sq_AL-edon-medium/espeak-ng-data",
            tokens="vits-piper-sq_AL-edon-medium/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Çdo fillim është i vështirë, por çdo fund është i bukur.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-sq_AL-edon-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx";
  config.model.vits.tokens = "vits-piper-sq_AL-edon-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-sq_AL-edon-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Çdo fillim është i vështirë, por çdo fund është i bukur.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-sq_AL-edon-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx".into()),
                tokens: Some("vits-piper-sq_AL-edon-medium/tokens.txt".into()),
                data_dir: Some("vits-piper-sq_AL-edon-medium/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Çdo fillim është i vështirë, por çdo fund është i bukur.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-sq_AL-edon-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx',
        tokens: 'vits-piper-sq_AL-edon-medium/tokens.txt',
        dataDir: 'vits-piper-sq_AL-edon-medium/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Çdo fillim është i vështirë, por çdo fund është i bukur.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx',
    tokens: 'vits-piper-sq_AL-edon-medium/tokens.txt',
    dataDir: 'vits-piper-sq_AL-edon-medium/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Çdo fillim është i vështirë, por çdo fund është i bukur.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-sq_AL-edon-medium/tokens.txt",
    dataDir: "vits-piper-sq_AL-edon-medium/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Çdo fillim është i vështirë, por çdo fund është i bukur."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-sq_AL-edon-medium/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-sq_AL-edon-medium/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Çdo fillim është i vështirë, por çdo fund është i bukur.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx",
        tokens = "vits-piper-sq_AL-edon-medium/tokens.txt",
        dataDir = "vits-piper-sq_AL-edon-medium/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Çdo fillim është i vështirë, por çdo fund është i bukur.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx");
    vits.setTokens("vits-piper-sq_AL-edon-medium/tokens.txt");
    vits.setDataDir("vits-piper-sq_AL-edon-medium/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Çdo fillim është i vështirë, por çdo fund është i bukur.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-sq_AL-edon-medium/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-sq_AL-edon-medium/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Çdo fillim është i vështirë, por çdo fund është i bukur.', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-sq_AL-edon-medium with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-sq_AL-edon-medium/sq_AL-edon-medium.onnx",
				Tokens:  "vits-piper-sq_AL-edon-medium/tokens.txt",
				DataDir: "vits-piper-sq_AL-edon-medium/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Çdo fillim është i vështirë, por çdo fund është i bukur."

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-eu_ES-antton-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-eu_ES-antton-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx";
  config.model.vits.tokens = "vits-piper-eu_ES-antton-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-eu_ES-antton-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Aberats izatea baino, izen ona hobe.";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-eu_ES-antton-medium.tar.bz2

You can use the following code to play with vits-piper-eu_ES-antton-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx",
            data_dir="vits-piper-eu_ES-antton-medium/espeak-ng-data",
            tokens="vits-piper-eu_ES-antton-medium/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Aberats izatea baino, izen ona hobe.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-eu_ES-antton-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx";
  config.model.vits.tokens = "vits-piper-eu_ES-antton-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-eu_ES-antton-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Aberats izatea baino, izen ona hobe.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-eu_ES-antton-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx".into()),
                tokens: Some("vits-piper-eu_ES-antton-medium/tokens.txt".into()),
                data_dir: Some("vits-piper-eu_ES-antton-medium/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Aberats izatea baino, izen ona hobe.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-eu_ES-antton-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx',
        tokens: 'vits-piper-eu_ES-antton-medium/tokens.txt',
        dataDir: 'vits-piper-eu_ES-antton-medium/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Aberats izatea baino, izen ona hobe.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx',
    tokens: 'vits-piper-eu_ES-antton-medium/tokens.txt',
    dataDir: 'vits-piper-eu_ES-antton-medium/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Aberats izatea baino, izen ona hobe.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-eu_ES-antton-medium/tokens.txt",
    dataDir: "vits-piper-eu_ES-antton-medium/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Aberats izatea baino, izen ona hobe."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-eu_ES-antton-medium/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-eu_ES-antton-medium/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Aberats izatea baino, izen ona hobe.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx",
        tokens = "vits-piper-eu_ES-antton-medium/tokens.txt",
        dataDir = "vits-piper-eu_ES-antton-medium/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Aberats izatea baino, izen ona hobe.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx");
    vits.setTokens("vits-piper-eu_ES-antton-medium/tokens.txt");
    vits.setDataDir("vits-piper-eu_ES-antton-medium/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Aberats izatea baino, izen ona hobe.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-eu_ES-antton-medium/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-eu_ES-antton-medium/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Aberats izatea baino, izen ona hobe.', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-antton-medium with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-eu_ES-antton-medium/eu_ES-antton-medium.onnx",
				Tokens:  "vits-piper-eu_ES-antton-medium/tokens.txt",
				DataDir: "vits-piper-eu_ES-antton-medium/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Aberats izatea baino, izen ona hobe."

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-eu_ES-maider-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-eu_ES-maider-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx";
  config.model.vits.tokens = "vits-piper-eu_ES-maider-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-eu_ES-maider-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Aberats izatea baino, izen ona hobe.";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-eu_ES-maider-medium.tar.bz2

You can use the following code to play with vits-piper-eu_ES-maider-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx",
            data_dir="vits-piper-eu_ES-maider-medium/espeak-ng-data",
            tokens="vits-piper-eu_ES-maider-medium/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Aberats izatea baino, izen ona hobe.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-eu_ES-maider-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx";
  config.model.vits.tokens = "vits-piper-eu_ES-maider-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-eu_ES-maider-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Aberats izatea baino, izen ona hobe.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-eu_ES-maider-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx".into()),
                tokens: Some("vits-piper-eu_ES-maider-medium/tokens.txt".into()),
                data_dir: Some("vits-piper-eu_ES-maider-medium/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Aberats izatea baino, izen ona hobe.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-eu_ES-maider-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx',
        tokens: 'vits-piper-eu_ES-maider-medium/tokens.txt',
        dataDir: 'vits-piper-eu_ES-maider-medium/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Aberats izatea baino, izen ona hobe.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx',
    tokens: 'vits-piper-eu_ES-maider-medium/tokens.txt',
    dataDir: 'vits-piper-eu_ES-maider-medium/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Aberats izatea baino, izen ona hobe.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-eu_ES-maider-medium/tokens.txt",
    dataDir: "vits-piper-eu_ES-maider-medium/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Aberats izatea baino, izen ona hobe."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-eu_ES-maider-medium/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-eu_ES-maider-medium/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Aberats izatea baino, izen ona hobe.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx",
        tokens = "vits-piper-eu_ES-maider-medium/tokens.txt",
        dataDir = "vits-piper-eu_ES-maider-medium/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Aberats izatea baino, izen ona hobe.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx");
    vits.setTokens("vits-piper-eu_ES-maider-medium/tokens.txt");
    vits.setDataDir("vits-piper-eu_ES-maider-medium/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Aberats izatea baino, izen ona hobe.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-eu_ES-maider-medium/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-eu_ES-maider-medium/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Aberats izatea baino, izen ona hobe.', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-eu_ES-maider-medium with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-eu_ES-maider-medium/eu_ES-maider-medium.onnx",
				Tokens:  "vits-piper-eu_ES-maider-medium/tokens.txt",
				DataDir: "vits-piper-eu_ES-maider-medium/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Aberats izatea baino, izen ona hobe."

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-supertonic-3-tts-int8-2026-05-11.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-supertonic-3-tts-int8-2026-05-11.tar.bz2

You can use the following code to play with supertonic-3

import sherpa_onnx
import soundfile as sf

tts_config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        supertonic=sherpa_onnx.OfflineTtsSupertonicModelConfig(
            duration_predictor="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
            text_encoder="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
            vector_estimator="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
            vocoder="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
            tts_json="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
            unicode_indexer="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
            voice_style="sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
        ),
        debug=False,
        num_threads=2,
        provider="cpu",
    ),
)

tts = sherpa_onnx.OfflineTts(tts_config)

gen_config = sherpa_onnx.GenerationConfig()
gen_config.sid = 0
gen_config.num_steps = 8
gen_config.speed = 1.0
gen_config.extra["lang"] = "bg"

audio = tts.generate("Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение", gen_config)

sf.write("test.wav", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with supertonic-3 with C API.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));
  config.model.supertonic.duration_predictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
  config.model.supertonic.text_encoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
  config.model.supertonic.vector_estimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
  config.model.supertonic.vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
  config.model.supertonic.tts_json = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
  config.model.supertonic.unicode_indexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
  config.model.supertonic.voice_style = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

  config.model.num_threads = 2;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 1;

  const char *text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.num_steps = 8;
  gen_cfg.speed = 1.0;
  gen_cfg.extra = "{\"lang\": \"bg\"}";

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg,
                                             ProgressCallback, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-supertonic.c. Then you can compile it with the following command:

gcc \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-supertonic \
  /tmp/test-supertonic.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-supertonic

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-supertonic.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with supertonic-3 with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;

  config.model.supertonic.duration_predictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
  config.model.supertonic.text_encoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
  config.model.supertonic.vector_estimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
  config.model.supertonic.vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
  config.model.supertonic.tts_json = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
  config.model.supertonic.unicode_indexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
  config.model.supertonic.voice_style = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

  config.model.num_threads = 2;

  // If you don't want to see debug messages, please set it to 0
  config.model.debug = 1;

  std::string filename = "./test.wav";
  std::string text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.num_steps = 8;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed
  gen_cfg.extra = R"({"lang": "bg"})";

  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake \
 -DSHERPA_ONNX_ENABLE_C_API=ON \
 -DCMAKE_BUILD_TYPE=Release \
 -DBUILD_SHARED_LIBS=ON \
 -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared \
 ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-supertonic.cc. Then you can compile it with the following command:

g++ \
  -std=c++17 \
  -I /tmp/sherpa-onnx/shared/include \
  -L /tmp/sherpa-onnx/shared/lib \
  -lsherpa-onnx-cxx-api \
  -lsherpa-onnx-c-api \
  -lonnxruntime \
  -o /tmp/test-supertonic \
  /tmp/test-supertonic.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-supertonic

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-supertonic.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with supertonic-3 with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsSupertonicModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            supertonic: OfflineTtsSupertonicModelConfig {
                duration_predictor: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx".into()),
                text_encoder: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx".into()),
                vector_estimator: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx".into()),
                vocoder: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx".into()),
                tts_json: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json".into()),
                unicode_indexer: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin".into()),
                voice_style: Some("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение";

    let gen_config = GenerationConfig {
        sid: 0,
        num_steps: 8,
        speed: 1.0,
        extra: Some(r#"{"lang": "bg"}"#.into()),
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with supertonic-3 with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      supertonic: {
        durationPredictor: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx',
        textEncoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx',
        vectorEstimator: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx',
        vocoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx',
        ttsJson: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json',
        unicodeIndexer: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin',
        voiceStyle: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin',
      },
      debug: true,
      numThreads: 2,
      provider: 'cpu',
    },
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  numSteps: 8,
  speed: 1.0,
  extra: {lang: 'bg'},
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with supertonic-3 with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final supertonic = sherpa_onnx.OfflineTtsSupertonicModelConfig(
    durationPredictor: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx',
    textEncoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx',
    vectorEstimator: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx',
    vocoder: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx',
    ttsJson: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json',
    unicodeIndexer: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin',
    voiceStyle: 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    supertonic: supertonic,
    numThreads: 2,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    numSteps: 8,
    speed: 1.0,
    extra: {'lang': 'bg'},
  );
  final audio = tts.generateWithConfig(text: 'Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with supertonic-3 with Swift API.

func run() {
  let supertonic = sherpaOnnxOfflineTtsSupertonicModelConfig(
    durationPredictor: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
    textEncoder: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
    vectorEstimator: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
    vocoder: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
    ttsJson: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
    unicodeIndexer: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
    voiceStyle: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(supertonic: supertonic)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.numSteps = 8
  genConfig.speed = 1.0
  genConfig.extra = ["lang": "bg"]

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with supertonic-3 with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Supertonic.DurationPredictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx";
config.Model.Supertonic.TextEncoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx";
config.Model.Supertonic.VectorEstimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx";
config.Model.Supertonic.Vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx";
config.Model.Supertonic.TtsJson = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json";
config.Model.Supertonic.UnicodeIndexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin";
config.Model.Supertonic.VoiceStyle = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin";

config.Model.NumThreads = 2;
config.Model.Debug = 1;
config.Model.Provider = "cpu";

var tts = new OfflineTts(config);
var text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.NumSteps = 8;
genConfig.Speed = 1.0f;
genConfig.Extra = "{\"lang\": \"bg\"}";

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with supertonic-3 with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      supertonic = OfflineTtsSupertonicModelConfig(
        durationPredictor = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
        textEncoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
        vectorEstimator = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
        vocoder = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
        ttsJson = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
        unicodeIndexer = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
        voiceStyle = "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
      ),
      numThreads = 2,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    numSteps = 8,
    speed = 1.0f,
    extra = mapOf("lang" to "bg"),
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with supertonic-3 with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var supertonic = new OfflineTtsSupertonicModelConfig();
    supertonic.setDurationPredictor("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx");
    supertonic.setTextEncoder("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx");
    supertonic.setVectorEstimator("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx");
    supertonic.setVocoder("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx");
    supertonic.setTtsJson("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json");
    supertonic.setUnicodeIndexer("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin");
    supertonic.setVoiceStyle("sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setSupertonic(supertonic);
    modelConfig.setNumThreads(2);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);

    var tts = new OfflineTts(config);
    var text = "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setNumSteps(8);
    genConfig.setSpeed(1.0f);
    genConfig.setExtra("{\"lang\": \"bg\"}");

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with supertonic-3 with Pascal API.

program test_supertonic;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Supertonic.DurationPredictor := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx';
  Config.Model.Supertonic.TextEncoder := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx';
  Config.Model.Supertonic.VectorEstimator := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx';
  Config.Model.Supertonic.Vocoder := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx';
  Config.Model.Supertonic.TtsJson := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json';
  Config.Model.Supertonic.UnicodeIndexer := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin';
  Config.Model.Supertonic.VoiceStyle := 'sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin';

  Config.Model.NumThreads := 2;
  Config.Model.Debug := True;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.NumSteps := 8;
  GenConfig.Speed := 1.0;
  GenConfig.Extra := '{"lang": "bg"}';

  Audio := Tts.GenerateWithConfig('Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with supertonic-3 with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Supertonic: sherpa.OfflineTtsSupertonicModelConfig{
				DurationPredictor: "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/duration_predictor.int8.onnx",
				TextEncoder:       "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/text_encoder.int8.onnx",
				VectorEstimator:   "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vector_estimator.int8.onnx",
				Vocoder:           "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/vocoder.int8.onnx",
				TtsJson:           "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/tts.json",
				UnicodeIndexer:    "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/unicode_indexer.bin",
				VoiceStyle:        "sherpa-onnx-supertonic-3-tts-int8-2026-05-11/voice.bin",
			},
			NumThreads: 2,
			Debug:      true,
		},
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Това е машина за преобразуване на текст в реч, използваща Kaldi от следващо поколение"

	genConfig := sherpa.GenerationConfig{
		Sid:       0,
		NumSteps:  8,
		Speed:     1.0,
		Extra:     `{"lang": "bg"}`,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_ona-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_ona-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Si vols estar ben servit, fes-te tu mateix el llit";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_ona-medium.tar.bz2

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx",
            data_dir="vits-piper-ca_ES-upc_ona-medium/espeak-ng-data",
            tokens="vits-piper-ca_ES-upc_ona-medium/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Si vols estar ben servit, fes-te tu mateix el llit",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_ona-medium/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Si vols estar ben servit, fes-te tu mateix el llit";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx".into()),
                tokens: Some("vits-piper-ca_ES-upc_ona-medium/tokens.txt".into()),
                data_dir: Some("vits-piper-ca_ES-upc_ona-medium/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Si vols estar ben servit, fes-te tu mateix el llit";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx',
        tokens: 'vits-piper-ca_ES-upc_ona-medium/tokens.txt',
        dataDir: 'vits-piper-ca_ES-upc_ona-medium/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Si vols estar ben servit, fes-te tu mateix el llit';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx',
    tokens: 'vits-piper-ca_ES-upc_ona-medium/tokens.txt',
    dataDir: 'vits-piper-ca_ES-upc_ona-medium/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Si vols estar ben servit, fes-te tu mateix el llit', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-ca_ES-upc_ona-medium/tokens.txt",
    dataDir: "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Si vols estar ben servit, fes-te tu mateix el llit"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-ca_ES-upc_ona-medium/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Si vols estar ben servit, fes-te tu mateix el llit";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx",
        tokens = "vits-piper-ca_ES-upc_ona-medium/tokens.txt",
        dataDir = "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Si vols estar ben servit, fes-te tu mateix el llit",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx");
    vits.setTokens("vits-piper-ca_ES-upc_ona-medium/tokens.txt");
    vits.setDataDir("vits-piper-ca_ES-upc_ona-medium/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Si vols estar ben servit, fes-te tu mateix el llit";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ca_ES-upc_ona-medium/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ca_ES-upc_ona-medium/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Si vols estar ben servit, fes-te tu mateix el llit', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-medium with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ca_ES-upc_ona-medium/ca_ES-upc_ona-medium.onnx",
				Tokens:  "vits-piper-ca_ES-upc_ona-medium/tokens.txt",
				DataDir: "vits-piper-ca_ES-upc_ona-medium/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Si vols estar ben servit, fes-te tu mateix el llit"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_ona-x_low.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_ona-x_low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Si vols estar ben servit, fes-te tu mateix el llit";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_ona-x_low.tar.bz2

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx",
            data_dir="vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data",
            tokens="vits-piper-ca_ES-upc_ona-x_low/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Si vols estar ben servit, fes-te tu mateix el llit",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_ona-x_low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Si vols estar ben servit, fes-te tu mateix el llit";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx".into()),
                tokens: Some("vits-piper-ca_ES-upc_ona-x_low/tokens.txt".into()),
                data_dir: Some("vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Si vols estar ben servit, fes-te tu mateix el llit";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx',
        tokens: 'vits-piper-ca_ES-upc_ona-x_low/tokens.txt',
        dataDir: 'vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Si vols estar ben servit, fes-te tu mateix el llit';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx',
    tokens: 'vits-piper-ca_ES-upc_ona-x_low/tokens.txt',
    dataDir: 'vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Si vols estar ben servit, fes-te tu mateix el llit', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx",
    lexicon: "",
    tokens: "vits-piper-ca_ES-upc_ona-x_low/tokens.txt",
    dataDir: "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Si vols estar ben servit, fes-te tu mateix el llit"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx";
config.Model.Vits.Tokens = "vits-piper-ca_ES-upc_ona-x_low/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Si vols estar ben servit, fes-te tu mateix el llit";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx",
        tokens = "vits-piper-ca_ES-upc_ona-x_low/tokens.txt",
        dataDir = "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Si vols estar ben servit, fes-te tu mateix el llit",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx");
    vits.setTokens("vits-piper-ca_ES-upc_ona-x_low/tokens.txt");
    vits.setDataDir("vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Si vols estar ben servit, fes-te tu mateix el llit";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ca_ES-upc_ona-x_low/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Si vols estar ben servit, fes-te tu mateix el llit', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_ona-x_low with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ca_ES-upc_ona-x_low/ca_ES-upc_ona-x_low.onnx",
				Tokens:  "vits-piper-ca_ES-upc_ona-x_low/tokens.txt",
				DataDir: "vits-piper-ca_ES-upc_ona-x_low/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Si vols estar ben servit, fes-te tu mateix el llit"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_pau-x_low.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_pau-x_low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "Si vols estar ben servit, fes-te tu mateix el llit";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-ca_ES-upc_pau-x_low.tar.bz2

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx",
            data_dir="vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data",
            tokens="vits-piper-ca_ES-upc_pau-x_low/tokens.txt",
        ),
        num_threads=1,
    ),
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="Si vols estar ben servit, fes-te tu mateix el llit",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx";
  config.model.vits.tokens = "vits-piper-ca_ES-upc_pau-x_low/tokens.txt";
  config.model.vits.data_dir = "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;

  std::string filename = "./test.wav";
  std::string text = "Si vols estar ben servit, fes-te tu mateix el llit";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx".into()),
                tokens: Some("vits-piper-ca_ES-upc_pau-x_low/tokens.txt".into()),
                data_dir: Some("vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "Si vols estar ben servit, fes-te tu mateix el llit";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx',
        tokens: 'vits-piper-ca_ES-upc_pau-x_low/tokens.txt',
        dataDir: 'vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = 'Si vols estar ben servit, fes-te tu mateix el llit';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx',
    tokens: 'vits-piper-ca_ES-upc_pau-x_low/tokens.txt',
    dataDir: 'vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: 'Si vols estar ben servit, fes-te tu mateix el llit', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx",
    lexicon: "",
    tokens: "vits-piper-ca_ES-upc_pau-x_low/tokens.txt",
    dataDir: "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data"
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "Si vols estar ben servit, fes-te tu mateix el llit"
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx";
config.Model.Vits.Tokens = "vits-piper-ca_ES-upc_pau-x_low/tokens.txt";
config.Model.Vits.DataDir = "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "Si vols estar ben servit, fes-te tu mateix el llit";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx",
        tokens = "vits-piper-ca_ES-upc_pau-x_low/tokens.txt",
        dataDir = "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data",
      ),
      numThreads = 1,
      debug = true,
    ),
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "Si vols estar ben servit, fes-te tu mateix el llit",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx");
    vits.setTokens("vits-piper-ca_ES-upc_pau-x_low/tokens.txt");
    vits.setDataDir("vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "Si vols estar ben servit, fes-te tu mateix el llit";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Pascal API.

program test_piper;

{$mode objfpc}

uses
  SysUtils,
  sherpa_onnx;

var
  Config: TSherpaOnnxOfflineTtsConfig;
  Tts: TSherpaOnnxOfflineTts;
  Audio: TSherpaOnnxGeneratedAudio;
  GenConfig: TSherpaOnnxGenerationConfig;

begin
  FillChar(Config, SizeOf(Config), 0);

  Config.Model.Vits.Model := 'vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx';
  Config.Model.Vits.Tokens := 'vits-piper-ca_ES-upc_pau-x_low/tokens.txt';
  Config.Model.Vits.DataDir := 'vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data';
  Config.Model.NumThreads := 1;
  Config.Model.Debug := True;
  Config.MaxNumSentences := 1;

  Tts := TSherpaOnnxOfflineTts.Create(@Config);

  GenConfig.Sid := 0;
  GenConfig.Speed := 1.0;
  GenConfig.SilenceScale := 0.2;

  Audio := Tts.GenerateWithConfig('Si vols estar ben servit, fes-te tu mateix el llit', @GenConfig, nil);

  WriteWave('./test.wav', Audio.Samples, Audio.N, Audio.SampleRate);

  WriteLn('Saved to ./test.wav');

  Audio.Free;
  Tts.Free;
end.

Please refer to the Pascal API documentation for more details.

You can use the following code to play with vits-piper-ca_ES-upc_pau-x_low with Go API.

package main

import (
	"fmt"
	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
)

func main() {
	config := sherpa.OfflineTtsConfig{
		Model: sherpa.OfflineTtsModelConfig{
			Vits: sherpa.OfflineTtsVitsModelConfig{
				Model:   "vits-piper-ca_ES-upc_pau-x_low/ca_ES-upc_pau-x_low.onnx",
				Tokens:  "vits-piper-ca_ES-upc_pau-x_low/tokens.txt",
				DataDir: "vits-piper-ca_ES-upc_pau-x_low/espeak-ng-data",
			},
			NumThreads: 1,
			Debug:      true,
		},
		MaxNumSentences: 1,
	}

	tts := sherpa.NewOfflineTts(&config)
	defer tts.Delete()

	text := "Si vols estar ben servit, fes-te tu mateix el llit"

	genConfig := sherpa.GenerationConfig{
		Sid:          0,
		Speed:        1.0,
		SilenceScale: 0.2,
	}

	audio := tts.GenerateWithConfig(text, &genConfig, nil)

	filename := "./test.wav"
	sherpa.WriteWave(filename, audio.Samples, audio.SampleRate)

	fmt.Printf("Saved to %s\n", filename)
}

Please refer to the Go API documentation for more details.

Model download address

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-zh_CN-chaowen-medium.tar.bz2

The following table shows the Android TTS Engine APK with this model for sherpa-onnx v1.13.2

If you don’t know what ABI is, you probably need to select arm64-v8a.

The source code for the APK can be found at

https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxTtsEngine

Please refer to the documentation for how to build the APK from source code.

More Android APKs can be found at

https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with C API.

#include <stdio.h>
#include <string.h>

#include "sherpa-onnx/c-api/c-api.h"

int main() {
  SherpaOnnxOfflineTtsConfig config;
  memset(&config, 0, sizeof(config));

  config.model.vits.model = "vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx";
  config.model.vits.lexicon = "vits-piper-zh_CN-chaowen-medium/lexicon.txt";
  config.model.vits.tokens = "vits-piper-zh_CN-chaowen-medium/tokens.txt";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;
  config.rule_fsts = "vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst";

  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);

  const char *text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.";

  SherpaOnnxGenerationConfig gen_cfg;
  memset(&gen_cfg, 0, sizeof(gen_cfg));
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0;

  const SherpaOnnxGeneratedAudio *audio =
      SherpaOnnxOfflineTtsGenerateWithConfig(tts, text, &gen_cfg, NULL, NULL);

  SherpaOnnxWriteWave(audio->samples, audio->n, audio->sample_rate,
                      "./test.wav");

  // You need to free the pointers to avoid memory leak in your app
  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
  SherpaOnnxDestroyOfflineTts(tts);

  printf("Saved to ./test.wav\n");

  return 0;
}

In the following, we describe how to compile and run the above C example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.c. Then you can compile it with the following command:

gcc   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.c

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

Assume you have installed sherpa-onnx via

pip install sherpa-onnx

and you have downloaded the model from

https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-zh_CN-chaowen-medium.tar.bz2

You can use the following code to play with vits-piper-zh_CN-chaowen-medium

import sherpa_onnx
import soundfile as sf

config = sherpa_onnx.OfflineTtsConfig(
    model=sherpa_onnx.OfflineTtsModelConfig(
        vits=sherpa_onnx.OfflineTtsVitsModelConfig(
            model="vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx",
            lexicon="vits-piper-zh_CN-chaowen-medium/lexicon.txt",
            tokens="vits-piper-zh_CN-chaowen-medium/tokens.txt",
        ),
        num_threads=1,
    ),
    rule_fsts="vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst",
)

if not config.validate():
    raise ValueError("Please check your config")

tts = sherpa_onnx.OfflineTts(config)
audio = tts.generate(text="某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.",
                     sid=0,
                     speed=1.0)

sf.write("test.mp3", audio.samples, samplerate=audio.sample_rate)

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with C++ API.

#include <cstdint>
#include <cstdio>
#include <string>

#include "sherpa-onnx/c-api/cxx-api.h"

static int32_t ProgressCallback(const float *samples, int32_t num_samples,
                                float progress, void *arg) {
  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
  // return 1 to continue generating
  // return 0 to stop generating
  return 1;
}

int32_t main(int32_t argc, char *argv[]) {
  using namespace sherpa_onnx::cxx; // NOLINT
  OfflineTtsConfig config;
  config.model.vits.model = "vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx";
  config.model.vits.lexicon = "vits-piper-zh_CN-chaowen-medium/lexicon.txt";
  config.model.vits.tokens = "vits-piper-zh_CN-chaowen-medium/tokens.txt";
  config.model.num_threads = 1;

  // If you want to see debug messages, please set it to 1
  config.model.debug = 0;
  config.rule_fsts = "vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst";

  std::string filename = "./test.wav";
  std::string text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.";

  auto tts = OfflineTts::Create(config);

  GenerationConfig gen_cfg;
  gen_cfg.sid = 0;
  gen_cfg.speed = 1.0; // larger -> faster in speech speed

#if 0
  // If you don't want to use a callback, then please enable this branch
  GeneratedAudio audio = tts.Generate(text, gen_cfg);
#else
  GeneratedAudio audio = tts.Generate(text, gen_cfg, ProgressCallback);
#endif

  WriteWave(filename, {audio.samples, audio.sample_rate});

  fprintf(stderr, "Input text is: %s\n", text.c_str());
  fprintf(stderr, "Speaker ID is: %d\n", gen_cfg.sid);
  fprintf(stderr, "Saved to: %s\n", filename.c_str());

  return 0;
}

In the following, we describe how to compile and run the above C++ example.

Use shared library (dynamic link)

cd /tmp
git clone https://github.com/k2-fsa/sherpa-onnx
cd sherpa-onnx
mkdir build-shared
cd build-shared

cmake  -DSHERPA_ONNX_ENABLE_C_API=ON  -DCMAKE_BUILD_TYPE=Release  -DBUILD_SHARED_LIBS=ON  -DCMAKE_INSTALL_PREFIX=/tmp/sherpa-onnx/shared  ..

make
make install

You can find required header file and library files inside /tmp/sherpa-onnx/shared.

Assume you have saved the above example file as /tmp/test-piper.cc. Then you can compile it with the following command:

g++   -std=c++17   -I /tmp/sherpa-onnx/shared/include   -L /tmp/sherpa-onnx/shared/lib   -lsherpa-onnx-cxx-api   -lsherpa-onnx-c-api   -lonnxruntime   -o /tmp/test-piper   /tmp/test-piper.cc

Now you can run

cd /tmp

# Assume you have downloaded the model and extracted it to /tmp
./test-piper

You probably need to run

# For Linux
export LD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$LD_LIBRARY_PATH

# For macOS
export DYLD_LIBRARY_PATH=/tmp/sherpa-onnx/shared/lib:$DYLD_LIBRARY_PATH

before you run /tmp/test-piper.

Use static library (static link)

Please see the documentation at

https://k2-fsa.github.io/sherpa/onnx/c-api/index.html

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with Rust API.

use sherpa_onnx::{
    GenerationConfig, OfflineTts, OfflineTtsConfig, OfflineTtsVitsModelConfig,
};

fn main() {
    let config = OfflineTtsConfig {
        model: sherpa_onnx::OfflineTtsModelConfig {
            vits: OfflineTtsVitsModelConfig {
                model: Some("vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx".into()),
                tokens: Some("vits-piper-zh_CN-chaowen-medium/tokens.txt".into()),
                lexicon: Some("vits-piper-zh_CN-chaowen-medium/lexicon.txt".into()),
                ..Default::default()
            },
            num_threads: 2,
            debug: false,
            ..Default::default()
        },
        rule_fsts: Some("vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst".into()),
        ..Default::default()
    };

    let tts = OfflineTts::create(&config).expect("Failed to create OfflineTts");

    println!("Sample rate: {}", tts.sample_rate());
    println!("Num speakers: {}", tts.num_speakers());

    let text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.";

    let gen_config = GenerationConfig {
        sid: 0,
        speed: 1.0,
        ..Default::default()
    };

    let audio = tts
        .generate_with_config(
            text,
            &gen_config,
            Some(|_samples: &[f32], progress: f32| -> bool {
                println!("Progress: {:.1}%", progress * 100.0);
                true
            }),
        )
        .expect("Generation failed");

    let filename = "./test.wav";
    if audio.save(filename) {
        println!("Saved to: {}", filename);
    } else {
        eprintln!("Failed to save {}", filename);
    }
}

Please refer to the Rust API documentation for how to build and run the above Rust example.

You need to install the sherpa-onnx-node npm package first:

npm install sherpa-onnx-node

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with the Node.js addon API.

const sherpa_onnx = require('sherpa-onnx-node');

function createOfflineTts() {
  const config = {
    model: {
      vits: {
        model: 'vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx',
        tokens: 'vits-piper-zh_CN-chaowen-medium/tokens.txt',
        lexicon: 'vits-piper-zh_CN-chaowen-medium/lexicon.txt',
      },
      debug: true,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
    ruleFsts: 'vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst',
  };
  return new sherpa_onnx.OfflineTts(config);
}

const tts = createOfflineTts();

const text = '某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 0,
  speed: 1.0,
  silenceScale: 0.2,
});

let start = Date.now();
const audio = tts.generate({text, generationConfig});
let stop = Date.now();
const elapsed_seconds = (stop - start) / 1000;
const duration = audio.samples.length / audio.sampleRate;
const real_time_factor = elapsed_seconds / duration;
console.log('Wave duration', duration.toFixed(3), 'seconds');
console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds');
console.log(
    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
    real_time_factor.toFixed(3));

const filename = 'test.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

console.log(`Saved to ${filename}`);

Please refer to the Node.js addon API documentation for more details.

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with Dart API.

import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;

void main() {
  final vits = sherpa_onnx.OfflineTtsVitsModelConfig(
    model: 'vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx',
    tokens: 'vits-piper-zh_CN-chaowen-medium/tokens.txt',
    dataDir: '',
  );

  final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
    vits: vits,
    numThreads: 1,
    debug: true,
  );
  final config = sherpa_onnx.OfflineTtsConfig(
    model: modelConfig,
    maxNumSenetences: 1,
  );

  final tts = sherpa_onnx.OfflineTts(config);
  final genConfig = sherpa_onnx.OfflineTtsGenerationConfig(
    sid: 0,
    speed: 1.0,
    silenceScale: 0.2,
  );
  final audio = tts.generateWithConfig(text: '某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.', config: genConfig);
  tts.free();

  sherpa_onnx.writeWave(
    filename: 'test.wav',
    samples: audio.samples,
    sampleRate: audio.sampleRate,
  );
  print('Saved to test.wav');
}

Please refer to the Dart API documentation for more details.

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with Swift API.

func run() {
  let vits = sherpaOnnxOfflineTtsVitsModelConfig(
    model: "vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx",
    lexicon: "",
    tokens: "vits-piper-zh_CN-chaowen-medium/tokens.txt",
    dataDir: ""
  )
  let modelConfig = sherpaOnnxOfflineTtsModelConfig(vits: vits)
  var ttsConfig = sherpaOnnxOfflineTtsConfig(model: modelConfig)

  let tts = SherpaOnnxOfflineTtsWrapper(config: &ttsConfig)

  let text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔."
  var genConfig = SherpaOnnxGenerationConfigSwift()
  genConfig.sid = 0
  genConfig.speed = 1.0
  genConfig.silenceScale = 0.2

  let audio = tts.generateWithConfig(text: text, config: genConfig, callback: nil, arg: nil)
  let filename = "test.wav"
  let ok = audio.save(filename: filename)
  if ok == 1 {
    print("Saved to \(filename)")
  } else {
    print("Failed to save \(filename)")
  }
}

@main
struct App {
  static func main() {
    run()
  }
}

Please refer to the Swift API documentation for more details.

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with C# API.

using SherpaOnnx;

var config = new OfflineTtsConfig();
config.Model.Vits.Model = "vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx";
config.Model.Vits.Tokens = "vits-piper-zh_CN-chaowen-medium/tokens.txt";
config.Model.Vits.Lexicon = "vits-piper-zh_CN-chaowen-medium/lexicon.txt";
config.Model.NumThreads = 1;
config.Model.Debug = 1;
config.Model.Provider = "cpu";
config.RuleFsts = "vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst";
config.MaxNumSentences = 1;

var tts = new OfflineTts(config);
var text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.";

OfflineTtsGenerationConfig genConfig = new OfflineTtsGenerationConfig();
genConfig.Sid = 0;
genConfig.Speed = 1.0f;
genConfig.SilenceScale = 0.2f;

var audio = tts.GenerateWithConfig(text, genConfig, null);
var ok = audio.SaveToWaveFile("./test.wav");

if (ok)
{
  Console.WriteLine("Saved to ./test.wav");
}
else
{
  Console.WriteLine("Failed to save ./test.wav");
}

Please refer to the C# API documentation for more details.

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with Kotlin API.

package com.k2fsa.sherpa.onnx

fun main() {
  var config = OfflineTtsConfig(
    model = OfflineTtsModelConfig(
      vits = OfflineTtsVitsModelConfig(
        model = "vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx",
        tokens = "vits-piper-zh_CN-chaowen-medium/tokens.txt",
        lexicon = "vits-piper-zh_CN-chaowen-medium/lexicon.txt",
      ),
      numThreads = 1,
      debug = true,
    ),
    ruleFsts = "vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst",
  )
  val tts = OfflineTts(config = config)
  val genConfig = GenerationConfig(
    sid = 0,
    speed = 1.0f,
    silenceScale = 0.2f,
  )
  val audio = tts.generateWithConfigAndCallback(
    text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.",
    config = genConfig,
    callback = ::callback,
  )
  audio.save(filename = "test.wav")
  tts.release()
  println("Saved to test.wav")
}

fun callback(samples: FloatArray): Int {
  // 1 means to continue
  // 0 means to stop
  return 1
}

Please refer to the Kotlin API documentation for more details.

You can use the following code to play with vits-piper-zh_CN-chaowen-medium with Java API.

import com.k2fsa.sherpa.onnx.*;

public class TtsDemo {
  public static void main(String[] args) {
    var vits = new OfflineTtsVitsModelConfig();
    vits.setModel("vits-piper-zh_CN-chaowen-medium/zh_CN-chaowen-medium.onnx");
    vits.setTokens("vits-piper-zh_CN-chaowen-medium/tokens.txt");
    vits.setLexicon("vits-piper-zh_CN-chaowen-medium/lexicon.txt");

    var modelConfig = new OfflineTtsModelConfig();
    modelConfig.setVits(vits);
    modelConfig.setNumThreads(1);
    modelConfig.setDebug(true);

    var config = new OfflineTtsConfig();
    config.setModel(modelConfig);
    config.setRuleFsts("vits-piper-zh_CN-chaowen-medium/phone.fst,vits-piper-zh_CN-chaowen-medium/date.fst,vits-piper-zh_CN-chaowen-medium/number.fst");
    config.setMaxNumSentences(1);

    var tts = new OfflineTts(config);
    var text = "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; 经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感受着生命的奇迹与温柔.";

    var genConfig = new GenerationConfig();
    genConfig.setSid(0);
    genConfig.setSpeed(1.0f);
    genConfig.setSilenceScale(0.2f);

    var audio = tts.generateWithConfigAndCallback(text, genConfig, (samples) -> {
      // 1 means to continue, 0 means to stop
      return 1;
    });

    audio.save("test.wav");
    tts.release();
    System.out.println("Saved to test.wav");
  }
}

Please refer to the Java API documentation for more details.