TTS: Kokoro (Chinese + English)

Generate speech with the Kokoro multi-language (Chinese+English, v1.0) model. This model supports mixed Chinese-English text with multiple speaker voices and both synchronous and asynchronous generation.

For model documentation, see Kokoro Chinese+English.

Source files

• Sync: test_tts_non_streaming_kokoro_zh_en.js

• Async: test_tts_non_streaming_kokoro_zh_en_async.js

Synchronous generation

// Copyright (c)  2025  Xiaomi Corporation
//
// Text-to-speech with the Kokoro multi-language (Chinese+English) model.
//
// Usage:
//   node tts_kokoro_zh_en.js
//
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',
        // Multiple lexicon files are separated by commas.
        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 =
    '中英文语音合成测试。This is generated by next generation Kaldi using Kokoro without Misaki. 你觉得中英文说的如何呢？';

const generationConfig = new sherpa_onnx.GenerationConfig({
  sid: 48,
  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-kokoro-zh-en.wav';
sherpa_onnx.writeWave(
    filename, {samples: audio.samples, sampleRate: audio.sampleRate});

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

Asynchronous generation

// Copyright (c)  2026  Xiaomi Corporation
//
// Asynchronous text-to-speech with the Kokoro Chinese+English model.
//
// Usage:
//   node tts_kokoro_zh_en_async.js
//
const sherpa_onnx = require('sherpa-onnx-node');

async 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: false,
      numThreads: 1,
      provider: 'cpu',
    },
    maxNumSentences: 1,
  };
  return await sherpa_onnx.OfflineTts.createAsync(config);
}

async function main() {
  const tts = await createOfflineTts();

  const text =
      '中英文语音合成测试。This is generated by next generation Kaldi using Kokoro without Misaki. 你觉得中英文说的如何呢？';

  const generationConfig = new sherpa_onnx.GenerationConfig({
    sid: 48,
    speed: 1.0,
    silenceScale: 0.2,
  });

  const start = Date.now();
  const audio = await tts.generateAsync({
    text,
    enableExternalBuffer: true,
    generationConfig,
    onProgress: ({samples, progress}) => {
      process.stdout.write(
          `Progress: ${(progress * 100).toFixed(1)}%, ` +
          `Samples: ${samples.length}\r`);
      return 1;
    },
  });

  console.log('');
  const 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-kokoro-zh-en-async.wav';
  sherpa_onnx.writeWave(
      filename, {samples: audio.samples, sampleRate: audio.sampleRate});
  console.log(`Saved to ${filename}`);
}

main().catch((err) => {
  console.error('Error:', err);
});

How to run

1. Install the package:

   npm install sherpa-onnx-node
   

2. Download the model:

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

3. Set the library path and run:

   # macOS
   export DYLD_LIBRARY_PATH=$(npm root)/sherpa-onnx-node/lib:$DYLD_LIBRARY_PATH
   
   # Linux
   export LD_LIBRARY_PATH=$(npm root)/sherpa-onnx-node/lib:$LD_LIBRARY_PATH
   
   # Choose one:
   node tts_kokoro_zh_en.js
   node tts_kokoro_zh_en_async.js
   

Notes

• This model uses the same kokoro config key as the English model, but adds a lexicon field with comma-separated lexicon files for each language (e.g., lexicon-us-en.txt,lexicon-zh.txt).

• sid: 48 selects a specific speaker voice. The multi-lang model has more speakers than the English-only model.

• The sync API uses new sherpa_onnx.OfflineTts(config) and tts.generate({text, generationConfig}).

• The async API uses OfflineTts.createAsync() and tts.generateAsync() with an onProgress callback.

• For English-only, see TTS: Kokoro (English).