Go

This page shows how to use X-ASR models with the Go API of sherpa-onnx. Two demos are given: one with a non-streaming model and one with a streaming model.

Requirements

  • Go 1.21 or above

  • CGO_ENABLED=1 (it is required by the native library)

Non-streaming X-ASR

Step 1. Create the project directory

mkdir x-asr-demo
cd x-asr-demo

Step 2. Create the build file

Save the following as go.mod:

module x-asr-demo

go 1.21

Step 3. Download the model

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03.tar.bz2
tar xf sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03.tar.bz2
rm sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03.tar.bz2

Step 4. Save the code

Save the following as main.go:

// This file shows how to use a non-streaming X-ASR model with sherpa-onnx
// in Go.
//
// Before you run it, please download the model:
//
//   wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03.tar.bz2
//   tar xf sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03.tar.bz2
package main

import (
     "fmt"
     "log"
     "time"

     sherpa "github.com/k2-fsa/sherpa-onnx-go/v2/sherpa_onnx"
)

func main() {
     modelDir := "./sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03"

     // Step 1: Create the recognizer from the X-ASR model files.
     config := sherpa.OfflineRecognizerConfig{}
     config.ModelConfig.Transducer.Encoder = modelDir + "/encoder-epoch-99-avg-1.int8.onnx"
     config.ModelConfig.Transducer.Decoder = modelDir + "/decoder-epoch-99-avg-1.onnx"
     config.ModelConfig.Transducer.Joiner = modelDir + "/joiner-epoch-99-avg-1.int8.onnx"
     config.ModelConfig.Tokens = modelDir + "/tokens.txt"
     config.ModelConfig.NumThreads = 2
     config.ModelConfig.Debug = 0
     config.DecodingMethod = "greedy_search"

     recognizer := sherpa.NewOfflineRecognizer(&config)
     defer sherpa.DeleteOfflineRecognizer(recognizer)

     // Step 2: Read the wave file you want to transcribe.
     filename := modelDir + "/test_wavs/0.wav"
     wave := sherpa.ReadWave(filename)
     if wave == nil {
             log.Fatalf("Failed to read %v", filename)
     }

     // Step 3: Create a stream, send the whole audio to it, and decode.
     start := time.Now()
     stream := sherpa.NewOfflineStream(recognizer)
     defer sherpa.DeleteOfflineStream(stream)
     stream.AcceptWaveform(wave.SampleRate, wave.Samples)
     recognizer.Decode(stream)

     // Step 4: Print the result. X-ASR punct models also add punctuations.
     fmt.Println(filename)
     elapsed := time.Since(start).Seconds()
     fmt.Println(stream.GetResult().Text)

     audioDuration := float64(len(wave.Samples)) / float64(wave.SampleRate)
     fmt.Printf("audio duration: %.3f s\n", audioDuration)
     fmt.Printf("elapsed seconds: %.3f s\n", elapsed)
     fmt.Printf("RTF = %.3f/%.3f = %.3f\n", elapsed, audioDuration, elapsed/audioDuration)
}

Step 5. Check the project layout

Your project directory should now look like this:

$ tree .
.
├── go.mod
├── main.go
└── sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03
    ├── bpe.model
    ├── decoder-epoch-99-avg-1.onnx
    ├── encoder-epoch-99-avg-1.int8.onnx
    ├── joiner-epoch-99-avg-1.int8.onnx
    ├── README.md
    ├── test_onnx.py
    ├── test_wavs
    │   ├── 0.wav
    │   ├── 1.wav
    │   ├── 2.wav
    │   └── 3.wav
    └── tokens.txt

3 directories, 13 files

$ ls -lh
total 16
-rw-r--r--@  1 fangjun  wheel    26B 19 Nov 12:23 go.mod
-rw-r--r--@  1 fangjun  wheel   2.1K 19 Nov 12:23 main.go
drwxr-xr-x@ 10 fangjun  wheel   320B 29 Jun 16:31 sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03

Step 6. Build and run

go mod tidy
go run .

Output

./sherpa-onnx-x-asr-zipformer-transducer-zh-en-punct-int8-2026-06-03/test_wavs/0.wav
昨天是 Monday, today is 礼拜二, the day after tomorrow 是星期三。
audio duration: 10.053 s
elapsed seconds: 0.108 s
RTF = 0.108/10.053 = 0.011

Note

The timing above was measured on a MacBook Pro (Apple M4) using 2 threads. Please increase the number of threads for a smaller RTF on your machine.

Streaming X-ASR

Step 1. Create the project directory

mkdir x-asr-demo
cd x-asr-demo

Step 2. Create the build file

Save the following as go.mod:

module x-asr-demo

go 1.21

Step 3. Download the model

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05.tar.bz2
tar xf sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05.tar.bz2
rm sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05.tar.bz2

Step 4. Save the code

Save the following as main.go:

// This file shows how to use a streaming X-ASR model with sherpa-onnx
// in Go.
//
// Before you run it, please download the model:
//
//   wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05.tar.bz2
//   tar xf sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05.tar.bz2
package main

import (
     "fmt"
     "log"
     "time"

     sherpa "github.com/k2-fsa/sherpa-onnx-go/v2/sherpa_onnx"
)

func main() {
     modelDir := "./sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05"

     // Step 1: Create the recognizer from the X-ASR model files.
     config := sherpa.OnlineRecognizerConfig{}
     config.ModelConfig.Transducer.Encoder = modelDir + "/encoder.int8.onnx"
     config.ModelConfig.Transducer.Decoder = modelDir + "/decoder.onnx"
     config.ModelConfig.Transducer.Joiner = modelDir + "/joiner.int8.onnx"
     config.ModelConfig.Tokens = modelDir + "/tokens.txt"
     config.ModelConfig.NumThreads = 2
     config.ModelConfig.Debug = 0
     config.DecodingMethod = "greedy_search"

     recognizer := sherpa.NewOnlineRecognizer(&config)
     defer sherpa.DeleteOnlineRecognizer(recognizer)

     // Step 2: Read the wave file you want to transcribe.
     filename := modelDir + "/test_wavs/0.wav"
     wave := sherpa.ReadWave(filename)
     if wave == nil {
             log.Fatalf("Failed to read %v", filename)
     }

     // Step 3: Create a stream and feed the audio in small chunks, just like
     // a microphone would do in a real-time application.
     start := time.Now()
     stream := sherpa.NewOnlineStream(recognizer)
     defer sherpa.DeleteOnlineStream(stream)

     // 0.2 second for 16 kHz audio
     chunk := 3200

     last := ""
     for i := 0; i < len(wave.Samples); i += chunk {
             end := i + chunk
             if end > len(wave.Samples) {
                     end = len(wave.Samples)
             }
             stream.AcceptWaveform(wave.SampleRate, wave.Samples[i:end])

             for recognizer.IsReady(stream) {
                     recognizer.Decode(stream)
             }

             // Print the latest partial result when it changes.
             result := recognizer.GetResult(stream)
             if result.Text != "" && result.Text != last {
                     last = result.Text
                     fmt.Println(result.Text)
             }
     }

     // Send some tail padding so the last words are flushed out.
     tail := make([]float32, int(0.5*float32(wave.SampleRate)))
     stream.AcceptWaveform(wave.SampleRate, tail)
     for recognizer.IsReady(stream) {
             recognizer.Decode(stream)
     }

     // Step 4: Print the final result.
     elapsed := time.Since(start).Seconds()
     fmt.Println("Final result:", recognizer.GetResult(stream).Text)

     audioDuration := float64(len(wave.Samples)) / float64(wave.SampleRate)
     fmt.Printf("audio duration: %.3f s\n", audioDuration)
     fmt.Printf("elapsed seconds: %.3f s\n", elapsed)
     fmt.Printf("RTF = %.3f/%.3f = %.3f\n", elapsed, audioDuration, elapsed/audioDuration)
}

Step 5. Check the project layout

Your project directory should now look like this:

$ tree .
.
├── go.mod
├── main.go
└── sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05
    ├── bpe.model
    ├── decoder.onnx
    ├── encoder.int8.onnx
    ├── joiner.int8.onnx
    ├── README.md
    ├── test_onnx.py
    ├── test_wavs
    │   ├── 0.wav
    │   ├── 1.wav
    │   ├── 2.wav
    │   └── 3.wav
    └── tokens.txt

3 directories, 13 files

$ ls -lh
total 16
-rw-r--r--@  1 fangjun  wheel    26B 19 Nov 12:23 go.mod
-rw-r--r--@  1 fangjun  wheel   2.7K 19 Nov 12:23 main.go
drwxr-xr-x@ 10 fangjun  wheel   320B 29 Jun 16:30 sherpa-onnx-x-asr-480ms-streaming-zipformer-transducer-zh-en-punct-int8-2026-06-05

Step 6. Build and run

go mod tidy
go run .

Output

昨
昨天是
昨天是 Mon
昨天是 Monday
昨天是 Monday,
昨天是 Monday, today
昨天是 Monday, today is 礼
昨天是 Monday, today is 礼拜二
昨天是 Monday, today is 礼拜二,
昨天是 Monday, today is 礼拜二, the day
昨天是 Monday, today is 礼拜二, the day after
昨天是 Monday, today is 礼拜二, the day after tomorro
昨天是 Monday, today is 礼拜二, the day after tomorrow
昨天是 Monday, today is 礼拜二, the day after tomorrow 是
Final result: 昨天是 Monday, today is 礼拜二, the day after tomorrow 是星期
audio duration: 10.053 s
elapsed seconds: 0.298 s
RTF = 0.298/10.053 = 0.030

Note

The timing above was measured on a MacBook Pro (Apple M4) using 2 threads. Please increase the number of threads for a smaller RTF on your machine.

Notes

If you see the error build constraints exclude all Go files, please run go env -w CGO_ENABLED=1 and try again.