Fangjun Kuang
Committed by GitHub

Add various languge bindings for Wenet non-streaming CTC models (#2584)

This PR adds support for Wenet non-streaming CTC models to sherpa-onnx by introducing the SherpaOnnxOfflineWenetCtcModelConfig struct and integrating it across all language bindings and APIs. The implementation follows the same pattern as other CTC model types like Zipformer CTC.

- Introduces SherpaOnnxOfflineWenetCtcModelConfig struct with a single model field for the ONNX model path
- Adds the new config to SherpaOnnxOfflineModelConfig and updates all language bindings (C++, Pascal, Kotlin, Java, Go, C#, Swift, JavaScript, etc.)
- Provides comprehensive examples and tests across all supported platforms and languages
正在显示 58 个修改的文件 包含 1393 行增加10 行删除
... ... @@ -70,6 +70,10 @@ popd
pushd non-streaming-asr
echo '----------Wenet CTC----------'
./run-wenet-ctc.sh
rm -rf sherpa-onnx-*
echo '----------Zipformer CTC----------'
./run-zipformer-ctc.sh
rm -rf sherpa-onnx-*
... ...
... ... @@ -27,6 +27,9 @@ rm -rf sherpa-onnx-nemo-*
cd ../offline-decode-files
./run-wenet-ctc.sh
rm -rf sherpa-onnx-*
./run-zipformer-ctc.sh
rm -rf sherpa-onnx-*
... ... @@ -108,6 +111,9 @@ cd ../keyword-spotting-from-files
./run.sh
cd ../online-decode-files
./run-t-one-ctc.sh
rm -rf sherpa-onnx-*
./run-transducer-itn.sh
rm -rf sherpa-onnx-*
... ...
... ... @@ -10,7 +10,16 @@ arch=$(node -p "require('os').arch()")
platform=$(node -p "require('os').platform()")
node_version=$(node -p "process.versions.node.split('.')[0]")
echo "----------streaming ASR T-one----------"
echo "----------non-streaming ASR Wenet CTC----------"
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
node ./test_asr_non_streaming_wenet_ctc.js
rm -rf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10
echo "----------streaming ASR T-one CTC----------"
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
tar xvf sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
rm sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
... ...
... ... @@ -9,6 +9,13 @@ git status
ls -lh
ls -lh node_modules
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
node ./test-offline-wenet-ctc.js
rm -rf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
tar xvf sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
rm sherpa-onnx-streaming-t-one-russian-2025-09-08.tar.bz2
... ...
... ... @@ -19,6 +19,9 @@ rm -fv *.wav *.onnx
ls -lh
rm -rf kitten-*
./run-wenet-ctc-asr.sh
rm -rf sherpa-onnx-*
./run-zipformer-ctc-asr.sh
rm -rf sherpa-onnx-zipformer-*
... ...
... ... @@ -75,6 +75,36 @@ jobs:
otool -L ./install/lib/libsherpa-onnx-c-api.dylib
fi
- name: Test Wenet CTC
shell: bash
run: |
name=wenet-ctc-c-api
gcc -o $name ./c-api-examples/$name.c \
-I ./build/install/include \
-L ./build/install/lib/ \
-l sherpa-onnx-c-api \
-l onnxruntime
ls -lh $name
if [[ ${{ matrix.os }} == ubuntu-latest || ${{ matrix.os }} == ubuntu-22.04-arm ]]; then
ldd ./$name
echo "----"
readelf -d ./$name
fi
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
./$name
rm $name
rm -rf sherpa-onnx-wenetspeech-*
- name: Test T-one
shell: bash
run: |
... ...
... ... @@ -78,6 +78,40 @@ jobs:
otool -L ./install/lib/libsherpa-onnx-cxx-api.dylib
fi
- name: Test Wenet CTC
shell: bash
run: |
name=wenet-ctc-cxx-api
g++ -std=c++17 -o $name ./cxx-api-examples/$name.cc \
-I ./build/install/include \
-L ./build/install/lib/ \
-l sherpa-onnx-cxx-api \
-l sherpa-onnx-c-api \
-l onnxruntime
ls -lh $name
if [[ ${{ matrix.os }} == ubuntu-latest || ${{ matrix.os }} == ubuntu-22.04-arm ]]; then
ls -lh ./$name
ldd ./$name
echo "----"
readelf -d ./$name
fi
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
echo "---"
export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
./$name
rm -rf sherpa-onnx-wenetspeech-*
rm -v ./$name
- name: Test T-one
shell: bash
run: |
... ...
... ... @@ -194,6 +194,10 @@ jobs:
go build
ls -lh
echo "Test Wenet CTC"
./run-wenet-ctc.sh
rm -rf sherpa-onnx-wenet*
echo "Test Zipformer CTC"
./run-zipformer-ctc.sh
rm -rf sherpa-onnx-zipformer-*
... ...
... ... @@ -151,3 +151,4 @@ kitten-nano-en-v0_1-fp16
vocab.json
*.so
sherpa-onnx-streaming-t-one-russian-2025-09-08
sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10
... ...
... ... @@ -80,6 +80,9 @@ target_link_libraries(moonshine-c-api sherpa-onnx-c-api)
add_executable(zipformer-c-api zipformer-c-api.c)
target_link_libraries(zipformer-c-api sherpa-onnx-c-api)
add_executable(wenet-ctc-c-api wenet-ctc-c-api.c)
target_link_libraries(wenet-ctc-c-api sherpa-onnx-c-api)
add_executable(streaming-zipformer-c-api streaming-zipformer-c-api.c)
target_link_libraries(streaming-zipformer-c-api sherpa-onnx-c-api)
... ...
// c-api-examples/wenet-ctc-c-api.c
//
// Copyright (c) 2025 Xiaomi Corporation
//
// This file demonstrates how to use non-streaming Wenet CTC model with
// sherpa-onnx's C API.
// clang-format off
//
// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
//
// clang-format on
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "sherpa-onnx/c-api/c-api.h"
int32_t main() {
// clang-format off
const char *wav_filename = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav";
const char *model = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx";
const char *tokens = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt";
// clang-format on
const char *provider = "cpu";
const SherpaOnnxWave *wave = SherpaOnnxReadWave(wav_filename);
if (wave == NULL) {
fprintf(stderr, "Failed to read %s\n", wav_filename);
return -1;
}
// Zipformer config
SherpaOnnxOfflineWenetCtcModelConfig wenet_ctc_config;
memset(&wenet_ctc_config, 0, sizeof(wenet_ctc_config));
wenet_ctc_config.model = model;
// Offline model config
SherpaOnnxOfflineModelConfig offline_model_config;
memset(&offline_model_config, 0, sizeof(offline_model_config));
offline_model_config.debug = 1;
offline_model_config.num_threads = 1;
offline_model_config.provider = provider;
offline_model_config.tokens = tokens;
offline_model_config.wenet_ctc = wenet_ctc_config;
// Recognizer config
SherpaOnnxOfflineRecognizerConfig recognizer_config;
memset(&recognizer_config, 0, sizeof(recognizer_config));
recognizer_config.decoding_method = "greedy_search";
recognizer_config.model_config = offline_model_config;
const SherpaOnnxOfflineRecognizer *recognizer =
SherpaOnnxCreateOfflineRecognizer(&recognizer_config);
if (recognizer == NULL) {
fprintf(stderr, "Please check your config!\n");
SherpaOnnxFreeWave(wave);
return -1;
}
const SherpaOnnxOfflineStream *stream =
SherpaOnnxCreateOfflineStream(recognizer);
SherpaOnnxAcceptWaveformOffline(stream, wave->sample_rate, wave->samples,
wave->num_samples);
SherpaOnnxDecodeOfflineStream(recognizer, stream);
const SherpaOnnxOfflineRecognizerResult *result =
SherpaOnnxGetOfflineStreamResult(stream);
fprintf(stderr, "Decoded text: %s\n", result->text);
SherpaOnnxDestroyOfflineRecognizerResult(result);
SherpaOnnxDestroyOfflineStream(stream);
SherpaOnnxDestroyOfflineRecognizer(recognizer);
SherpaOnnxFreeWave(wave);
return 0;
}
... ...
... ... @@ -30,6 +30,9 @@ target_link_libraries(moonshine-cxx-api sherpa-onnx-cxx-api)
add_executable(sense-voice-cxx-api ./sense-voice-cxx-api.cc)
target_link_libraries(sense-voice-cxx-api sherpa-onnx-cxx-api)
add_executable(wenet-ctc-cxx-api ./wenet-ctc-cxx-api.cc)
target_link_libraries(wenet-ctc-cxx-api sherpa-onnx-cxx-api)
add_executable(nemo-canary-cxx-api ./nemo-canary-cxx-api.cc)
target_link_libraries(nemo-canary-cxx-api sherpa-onnx-cxx-api)
... ... @@ -46,6 +49,15 @@ if(SHERPA_ONNX_ENABLE_PORTAUDIO)
portaudio_static
)
add_executable(wenet-ctc-simulate-streaming-microphone-cxx-api
./wenet-ctc-simulate-streaming-microphone-cxx-api.cc
${CMAKE_CURRENT_LIST_DIR}/../sherpa-onnx/csrc/microphone.cc
)
target_link_libraries(wenet-ctc-simulate-streaming-microphone-cxx-api
sherpa-onnx-cxx-api
portaudio_static
)
add_executable(parakeet-tdt-simulate-streaming-microphone-cxx-api
./parakeet-tdt-simulate-streaming-microphone-cxx-api.cc
${CMAKE_CURRENT_LIST_DIR}/../sherpa-onnx/csrc/microphone.cc
... ...
// cxx-api-examples/wenet-cxx-api.cc
// Copyright (c) 2025 Xiaomi Corporation
//
// This file demonstrates how to use Wenet CTC with sherpa-onnx's C++ API.
//
// clang-format off
//
// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
//
// clang-format on
#include <chrono> // NOLINT
#include <iostream>
#include <string>
#include "sherpa-onnx/c-api/cxx-api.h"
int32_t main() {
using namespace sherpa_onnx::cxx; // NOLINT
OfflineRecognizerConfig config;
// clang-format off
config.model_config.wenet_ctc.model = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx";
config.model_config.tokens = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt";
config.model_config.num_threads = 1;
std::cout << "Loading model\n";
OfflineRecognizer recognizer = OfflineRecognizer::Create(config);
if (!recognizer.Get()) {
std::cerr << "Please check your config\n";
return -1;
}
std::cout << "Loading model done\n";
std::string wave_filename = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav";
// clang-format on
Wave wave = ReadWave(wave_filename);
if (wave.samples.empty()) {
std::cerr << "Failed to read: '" << wave_filename << "'\n";
return -1;
}
std::cout << "Start recognition\n";
const auto begin = std::chrono::steady_clock::now();
OfflineStream stream = recognizer.CreateStream();
stream.AcceptWaveform(wave.sample_rate, wave.samples.data(),
wave.samples.size());
recognizer.Decode(&stream);
OfflineRecognizerResult result = recognizer.GetResult(&stream);
const auto end = std::chrono::steady_clock::now();
const float elapsed_seconds =
std::chrono::duration_cast<std::chrono::milliseconds>(end - begin)
.count() /
1000.;
float duration = wave.samples.size() / static_cast<float>(wave.sample_rate);
float rtf = elapsed_seconds / duration;
std::cout << "text: " << result.text << "\n";
printf("Number of threads: %d\n", config.model_config.num_threads);
printf("Duration: %.3fs\n", duration);
printf("Elapsed seconds: %.3fs\n", elapsed_seconds);
printf("(Real time factor) RTF = %.3f / %.3f = %.3f\n", elapsed_seconds,
duration, rtf);
return 0;
}
... ...
// cxx-api-examples/wenet-ctc-simulate-streaming-microphone-cxx-api.cc
// Copyright (c) 2025 Xiaomi Corporation
//
// This file demonstrates how to use Wenet CTC with sherpa-onnx's C++ API
// for streaming speech recognition from a microphone.
//
// clang-format off
//
//
// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
// rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
//
// clang-format on
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <chrono> // NOLINT
#include <condition_variable> // NOLINT
#include <iostream>
#include <mutex> // NOLINT
#include <queue>
#include <vector>
#include "portaudio.h" // NOLINT
#include "sherpa-display.h" // NOLINT
#include "sherpa-onnx/c-api/cxx-api.h"
#include "sherpa-onnx/csrc/microphone.h"
std::queue<std::vector<float>> samples_queue;
std::condition_variable condition_variable;
std::mutex mutex;
bool stop = false;
static void Handler(int32_t /*sig*/) {
stop = true;
condition_variable.notify_one();
fprintf(stderr, "\nCaught Ctrl + C. Exiting...\n");
}
static int32_t RecordCallback(const void *input_buffer,
void * /*output_buffer*/,
unsigned long frames_per_buffer, // NOLINT
const PaStreamCallbackTimeInfo * /*time_info*/,
PaStreamCallbackFlags /*status_flags*/,
void * /*user_data*/) {
std::lock_guard<std::mutex> lock(mutex);
samples_queue.emplace(
reinterpret_cast<const float *>(input_buffer),
reinterpret_cast<const float *>(input_buffer) + frames_per_buffer);
condition_variable.notify_one();
return stop ? paComplete : paContinue;
}
static sherpa_onnx::cxx::VoiceActivityDetector CreateVad() {
using namespace sherpa_onnx::cxx; // NOLINT
VadModelConfig config;
config.silero_vad.model = "./silero_vad.onnx";
config.silero_vad.threshold = 0.5;
config.silero_vad.min_silence_duration = 0.1;
config.silero_vad.min_speech_duration = 0.25;
config.silero_vad.max_speech_duration = 8;
config.sample_rate = 16000;
config.debug = false;
VoiceActivityDetector vad = VoiceActivityDetector::Create(config, 20);
if (!vad.Get()) {
std::cerr << "Failed to create VAD. Please check your config\n";
exit(-1);
}
return vad;
}
static sherpa_onnx::cxx::OfflineRecognizer CreateOfflineRecognizer() {
using namespace sherpa_onnx::cxx; // NOLINT
OfflineRecognizerConfig config;
// clang-format off
config.model_config.wenet_ctc.model = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx";
config.model_config.tokens = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt";
// clang-format on
config.model_config.num_threads = 2;
config.model_config.debug = false;
std::cout << "Loading model\n";
OfflineRecognizer recognizer = OfflineRecognizer::Create(config);
if (!recognizer.Get()) {
std::cerr << "Please check your config\n";
exit(-1);
}
std::cout << "Loading model done\n";
return recognizer;
}
int32_t main() {
signal(SIGINT, Handler);
using namespace sherpa_onnx::cxx; // NOLINT
auto vad = CreateVad();
auto recognizer = CreateOfflineRecognizer();
sherpa_onnx::Microphone mic;
PaDeviceIndex num_devices = Pa_GetDeviceCount();
if (num_devices == 0) {
std::cerr << " If you are using Linux, please try "
"./build/bin/sense-voice-simulate-streaming-alsa-cxx-api\n";
return -1;
}
int32_t device_index = Pa_GetDefaultInputDevice();
const char *pDeviceIndex = std::getenv("SHERPA_ONNX_MIC_DEVICE");
if (pDeviceIndex) {
fprintf(stderr, "Use specified device: %s\n", pDeviceIndex);
device_index = atoi(pDeviceIndex);
}
mic.PrintDevices(device_index);
float mic_sample_rate = 16000;
const char *sample_rate_str = std::getenv("SHERPA_ONNX_MIC_SAMPLE_RATE");
if (sample_rate_str) {
fprintf(stderr, "Use sample rate %f for mic\n", mic_sample_rate);
mic_sample_rate = atof(sample_rate_str);
}
float sample_rate = 16000;
LinearResampler resampler;
if (mic_sample_rate != sample_rate) {
float min_freq = std::min(mic_sample_rate, sample_rate);
float lowpass_cutoff = 0.99 * 0.5 * min_freq;
int32_t lowpass_filter_width = 6;
resampler = LinearResampler::Create(mic_sample_rate, sample_rate,
lowpass_cutoff, lowpass_filter_width);
}
if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
nullptr)) {
std::cerr << "Failed to open microphone device\n";
return -1;
}
int32_t window_size = 512; // samples, please don't change
int32_t offset = 0;
std::vector<float> buffer;
bool speech_started = false;
auto started_time = std::chrono::steady_clock::now();
SherpaDisplay display;
std::cout << "Started! Please speak\n";
while (!stop) {
{
std::unique_lock<std::mutex> lock(mutex);
while (samples_queue.empty() && !stop) {
condition_variable.wait(lock);
}
if (stop) {
break;
}
const auto &s = samples_queue.front();
if (!resampler.Get()) {
buffer.insert(buffer.end(), s.begin(), s.end());
} else {
auto resampled = resampler.Resample(s.data(), s.size(), false);
buffer.insert(buffer.end(), resampled.begin(), resampled.end());
}
samples_queue.pop();
}
for (; offset + window_size < buffer.size(); offset += window_size) {
vad.AcceptWaveform(buffer.data() + offset, window_size);
if (!speech_started && vad.IsDetected()) {
speech_started = true;
started_time = std::chrono::steady_clock::now();
}
}
if (!speech_started) {
if (buffer.size() > 10 * window_size) {
offset -= buffer.size() - 10 * window_size;
buffer = {buffer.end() - 10 * window_size, buffer.end()};
}
}
auto current_time = std::chrono::steady_clock::now();
const float elapsed_seconds =
std::chrono::duration_cast<std::chrono::milliseconds>(current_time -
started_time)
.count() /
1000.;
if (speech_started && elapsed_seconds > 0.2) {
OfflineStream stream = recognizer.CreateStream();
stream.AcceptWaveform(sample_rate, buffer.data(), buffer.size());
recognizer.Decode(&stream);
OfflineRecognizerResult result = recognizer.GetResult(&stream);
display.UpdateText(result.text);
display.Display();
started_time = std::chrono::steady_clock::now();
}
while (!vad.IsEmpty()) {
auto segment = vad.Front();
vad.Pop();
OfflineStream stream = recognizer.CreateStream();
stream.AcceptWaveform(sample_rate, segment.samples.data(),
segment.samples.size());
recognizer.Decode(&stream);
OfflineRecognizerResult result = recognizer.GetResult(&stream);
display.UpdateText(result.text);
display.FinalizeCurrentSentence();
display.Display();
buffer.clear();
offset = 0;
speech_started = false;
}
}
return 0;
}
... ...
// Copyright (c) 2025 Xiaomi Corporation
import 'dart:io';
import 'package:args/args.dart';
import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
import './init.dart';
void main(List<String> arguments) async {
await initSherpaOnnx();
final parser = ArgParser()
..addOption('model', help: 'Path to the Wenet CTC model')
..addOption('tokens', help: 'Path to tokens.txt')
..addOption('input-wav', help: 'Path to input.wav to transcribe');
final res = parser.parse(arguments);
if (res['model'] == null ||
res['tokens'] == null ||
res['input-wav'] == null) {
print(parser.usage);
exit(1);
}
final model = res['model'] as String;
final tokens = res['tokens'] as String;
final inputWav = res['input-wav'] as String;
final wenetCtc = sherpa_onnx.OfflineWenetCtcModelConfig(model: model);
final modelConfig = sherpa_onnx.OfflineModelConfig(
wenetCtc: wenetCtc,
tokens: tokens,
debug: true,
numThreads: 1,
);
final config = sherpa_onnx.OfflineRecognizerConfig(model: modelConfig);
final recognizer = sherpa_onnx.OfflineRecognizer(config);
final waveData = sherpa_onnx.readWave(inputWav);
final stream = recognizer.createStream();
stream.acceptWaveform(
samples: waveData.samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
print(result.text);
stream.free();
recognizer.free();
}
... ...
#!/usr/bin/env bash
set -ex
dart pub get
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
dart run \
./bin/wenet-ctc.dart \
--model ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx \
--tokens ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt \
--input-wav ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav
... ...
... ... @@ -84,6 +84,9 @@ class OfflineDecodeFiles
[Option("telespeech-ctc", Required = false, HelpText = "Path to model.onnx. Used only for TeleSpeech CTC models")]
public string TeleSpeechCtc { get; set; } = string.Empty;
[Option("wenet-ctc", Required = false, HelpText = "Path to model.onnx. Used only for Wenet CTC models")]
public string WenetCtc { get; set; } = string.Empty;
[Option("sense-voice-model", Required = false, HelpText = "Path to model.onnx. Used only for SenseVoice CTC models")]
public string SenseVoiceModel { get; set; } = string.Empty;
... ... @@ -251,6 +254,10 @@ to download pre-trained Tdnn models.
{
config.ModelConfig.TeleSpeechCtc = options.TeleSpeechCtc;
}
else if (!string.IsNullOrEmpty(options.WenetCtc))
{
config.ModelConfig.WenetCtc.Model = options.WenetCtc;
}
else if (!string.IsNullOrEmpty(options.WhisperEncoder))
{
config.ModelConfig.Whisper.Encoder = options.WhisperEncoder;
... ...
#!/usr/bin/env bash
set -ex
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
dotnet run \
--wenet-ctc=./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx \
--tokens=./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt \
--files ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav
... ...
... ... @@ -125,6 +125,27 @@ class OfflineZipformerCtcModelConfig {
final String model;
}
class OfflineWenetCtcModelConfig {
const OfflineWenetCtcModelConfig({this.model = ''});
factory OfflineWenetCtcModelConfig.fromJson(Map<String, dynamic> json) {
return OfflineWenetCtcModelConfig(
model: json['model'] as String? ?? '',
);
}
@override
String toString() {
return 'OfflineWenetCtcModelConfig(model: $model)';
}
Map<String, dynamic> toJson() => {
'model': model,
};
final String model;
}
class OfflineWhisperModelConfig {
const OfflineWhisperModelConfig(
{this.encoder = '',
... ... @@ -349,6 +370,7 @@ class OfflineModelConfig {
this.dolphin = const OfflineDolphinModelConfig(),
this.zipformerCtc = const OfflineZipformerCtcModelConfig(),
this.canary = const OfflineCanaryModelConfig(),
this.wenetCtc = const OfflineWenetCtcModelConfig(),
required this.tokens,
this.numThreads = 1,
this.debug = true,
... ... @@ -405,6 +427,10 @@ class OfflineModelConfig {
? OfflineCanaryModelConfig.fromJson(
json['canary'] as Map<String, dynamic>)
: const OfflineCanaryModelConfig(),
wenetCtc: json['wenetCtc'] != null
? OfflineWenetCtcModelConfig.fromJson(
json['wenetCtc'] as Map<String, dynamic>)
: const OfflineWenetCtcModelConfig(),
tokens: json['tokens'] as String,
numThreads: json['numThreads'] as int? ?? 1,
debug: json['debug'] as bool? ?? true,
... ... @@ -418,7 +444,7 @@ class OfflineModelConfig {
@override
String toString() {
return 'OfflineModelConfig(transducer: $transducer, paraformer: $paraformer, nemoCtc: $nemoCtc, whisper: $whisper, tdnn: $tdnn, senseVoice: $senseVoice, moonshine: $moonshine, fireRedAsr: $fireRedAsr, dolphin: $dolphin, zipformerCtc: $zipformerCtc, canary: $canary, tokens: $tokens, numThreads: $numThreads, debug: $debug, provider: $provider, modelType: $modelType, modelingUnit: $modelingUnit, bpeVocab: $bpeVocab, telespeechCtc: $telespeechCtc)';
return 'OfflineModelConfig(transducer: $transducer, paraformer: $paraformer, nemoCtc: $nemoCtc, whisper: $whisper, tdnn: $tdnn, senseVoice: $senseVoice, moonshine: $moonshine, fireRedAsr: $fireRedAsr, dolphin: $dolphin, zipformerCtc: $zipformerCtc, canary: $canary, wenetCtc: $wenetCtc, tokens: $tokens, numThreads: $numThreads, debug: $debug, provider: $provider, modelType: $modelType, modelingUnit: $modelingUnit, bpeVocab: $bpeVocab, telespeechCtc: $telespeechCtc)';
}
Map<String, dynamic> toJson() => {
... ... @@ -433,6 +459,7 @@ class OfflineModelConfig {
'dolphin': dolphin.toJson(),
'zipformerCtc': zipformerCtc.toJson(),
'canary': canary.toJson(),
'wenetCtc': wenetCtc.toJson(),
'tokens': tokens,
'numThreads': numThreads,
'debug': debug,
... ... @@ -454,6 +481,7 @@ class OfflineModelConfig {
final OfflineDolphinModelConfig dolphin;
final OfflineZipformerCtcModelConfig zipformerCtc;
final OfflineCanaryModelConfig canary;
final OfflineWenetCtcModelConfig wenetCtc;
final String tokens;
final int numThreads;
... ... @@ -690,6 +718,8 @@ class OfflineRecognizer {
c.ref.model.canary.tgtLang = config.model.canary.tgtLang.toNativeUtf8();
c.ref.model.canary.usePnc = config.model.canary.usePnc ? 1 : 0;
c.ref.model.wenetCtc.model = config.model.wenetCtc.model.toNativeUtf8();
c.ref.model.tokens = config.model.tokens.toNativeUtf8();
c.ref.model.numThreads = config.model.numThreads;
... ... @@ -736,6 +766,7 @@ class OfflineRecognizer {
calloc.free(c.ref.model.modelType);
calloc.free(c.ref.model.provider);
calloc.free(c.ref.model.tokens);
calloc.free(c.ref.model.wenetCtc.model);
calloc.free(c.ref.model.canary.tgtLang);
calloc.free(c.ref.model.canary.srcLang);
calloc.free(c.ref.model.canary.decoder);
... ...
... ... @@ -281,6 +281,10 @@ final class SherpaOnnxOfflineZipformerCtcModelConfig extends Struct {
external Pointer<Utf8> model;
}
final class SherpaOnnxOfflineWenetCtcModelConfig extends Struct {
external Pointer<Utf8> model;
}
final class SherpaOnnxOfflineWhisperModelConfig extends Struct {
external Pointer<Utf8> encoder;
external Pointer<Utf8> decoder;
... ... @@ -360,6 +364,7 @@ final class SherpaOnnxOfflineModelConfig extends Struct {
external SherpaOnnxOfflineDolphinModelConfig dolphin;
external SherpaOnnxOfflineZipformerCtcModelConfig zipformerCtc;
external SherpaOnnxOfflineCanaryModelConfig canary;
external SherpaOnnxOfflineWenetCtcModelConfig wenetCtc;
}
final class SherpaOnnxOfflineRecognizerConfig extends Struct {
... ...
... ... @@ -29,6 +29,7 @@ func main() {
flag.StringVar(&config.ModelConfig.NemoCTC.Model, "nemo-ctc", "", "Path to the NeMo CTC model")
flag.StringVar(&config.ModelConfig.ZipformerCtc.Model, "zipformer-ctc", "", "Path to the Zipformer CTC model")
flag.StringVar(&config.ModelConfig.WenetCtc.Model, "wenet-ctc", "", "Path to the Wenet CTC model")
flag.StringVar(&config.ModelConfig.Dolphin.Model, "dolphin-model", "", "Path to the Dolphin CTC model")
... ...
#!/usr/bin/env bash
set -ex
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
go mod tidy
go build
./non-streaming-decode-files \
--wenet-ctc ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx \
--tokens ./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt \
--debug 0 \
./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav
... ...
... ... @@ -14,8 +14,9 @@ export { Samples,
OfflineNemoEncDecCtcModelConfig,
OfflineWhisperModelConfig,
OfflineTdnnModelConfig,
OfflineSenseVoiceModelConfig,
OfflineMoonshineModelConfig,
OfflineSenseVoiceModelConfig,
OfflineWenetCtcModelConfig,
OfflineZipformerCtcModelConfig,
OfflineModelConfig,
OfflineLMConfig,
... ...
... ... @@ -61,6 +61,22 @@ GetOfflineZipformerCtcModelConfig(Napi::Object obj) {
return c;
}
static SherpaOnnxOfflineWenetCtcModelConfig GetOfflineWenetCtcModelConfig(
Napi::Object obj) {
SherpaOnnxOfflineWenetCtcModelConfig c;
memset(&c, 0, sizeof(c));
if (!obj.Has("wenetCtc") || !obj.Get("wenetCtc").IsObject()) {
return c;
}
Napi::Object o = obj.Get("wenetCtc").As<Napi::Object>();
SHERPA_ONNX_ASSIGN_ATTR_STR(model, model);
return c;
}
static SherpaOnnxOfflineDolphinModelConfig GetOfflineDolphinModelConfig(
Napi::Object obj) {
SherpaOnnxOfflineDolphinModelConfig c;
... ... @@ -225,6 +241,7 @@ static SherpaOnnxOfflineModelConfig GetOfflineModelConfig(Napi::Object obj) {
c.dolphin = GetOfflineDolphinModelConfig(o);
c.zipformer_ctc = GetOfflineZipformerCtcModelConfig(o);
c.canary = GetOfflineCanaryModelConfig(o);
c.wenet_ctc = GetOfflineWenetCtcModelConfig(o);
SHERPA_ONNX_ASSIGN_ATTR_STR(tokens, tokens);
SHERPA_ONNX_ASSIGN_ATTR_INT32(num_threads, numThreads);
... ... @@ -317,6 +334,8 @@ static void FreeConfig(const SherpaOnnxOfflineRecognizerConfig &c) {
SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.src_lang);
SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.tgt_lang);
SHERPA_ONNX_DELETE_C_STR(c.model_config.wenet_ctc.model);
SHERPA_ONNX_DELETE_C_STR(c.model_config.tokens);
SHERPA_ONNX_DELETE_C_STR(c.model_config.provider);
SHERPA_ONNX_DELETE_C_STR(c.model_config.model_type);
... ...
... ... @@ -60,6 +60,10 @@ export class OfflineZipformerCtcModelConfig {
public model: string = '';
}
export class OfflineWenetCtcModelConfig {
public model: string = '';
}
export class OfflineWhisperModelConfig {
public encoder: string = '';
public decoder: string = '';
... ... @@ -112,6 +116,7 @@ export class OfflineModelConfig {
public dolphin: OfflineDolphinModelConfig = new OfflineDolphinModelConfig();
public zipformerCtc: OfflineZipformerCtcModelConfig = new OfflineZipformerCtcModelConfig();
public canary: OfflineCanaryModelConfig = new OfflineCanaryModelConfig();
public wenetCtc: OfflineWenetCtcModelConfig = new OfflineWenetCtcModelConfig();
}
export class OfflineLMConfig {
... ...
// Copyright 2025 Xiaomi Corporation
// This file shows how to use an offline Wenet CTC model,
// i.e., non-streaming Wenet CTC model,
// to decode files.
import com.k2fsa.sherpa.onnx.*;
public class NonStreamingDecodeFileWenetCtc {
public static void main(String[] args) {
// please refer to
// https://k2-fsa.github.io/sherpa/onnx/sense-voice/index.html
// to download model files
String model =
"sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx";
String tokens =
"sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt";
String waveFilename =
"sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav";
WaveReader reader = new WaveReader(waveFilename);
OfflineWenetCtcModelConfig wenetCtc =
OfflineWenetCtcModelConfig.builder().setModel(model).build();
OfflineModelConfig modelConfig =
OfflineModelConfig.builder()
.setWenetCtc(wenetCtc)
.setTokens(tokens)
.setNumThreads(1)
.setDebug(true)
.build();
OfflineRecognizerConfig config =
OfflineRecognizerConfig.builder()
.setOfflineModelConfig(modelConfig)
.setDecodingMethod("greedy_search")
.build();
OfflineRecognizer recognizer = new OfflineRecognizer(config);
OfflineStream stream = recognizer.createStream();
stream.acceptWaveform(reader.getSamples(), reader.getSampleRate());
recognizer.decode(stream);
String text = recognizer.getResult(stream).getText();
System.out.printf("filename:%s\nresult:%s\n", waveFilename, text);
stream.release();
recognizer.release();
}
}
... ...
#!/usr/bin/env bash
set -ex
if [[ ! -f ../build/lib/libsherpa-onnx-jni.dylib && ! -f ../build/lib/libsherpa-onnx-jni.so ]]; then
mkdir -p ../build
pushd ../build
cmake \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
-DSHERPA_ONNX_ENABLE_JNI=ON \
..
make -j4
ls -lh lib
popd
fi
if [ ! -f ../sherpa-onnx/java-api/build/sherpa-onnx.jar ]; then
pushd ../sherpa-onnx/java-api
make
popd
fi
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
java \
-Djava.library.path=$PWD/../build/lib \
-cp ../sherpa-onnx/java-api/build/sherpa-onnx.jar \
NonStreamingDecodeFileWenetCtc.java
... ...
... ... @@ -489,8 +489,30 @@ function testOfflineNeMoCanary() {
java -Djava.library.path=../build/lib -jar $out_filename
}
function testOfflineWenetCtc() {
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
out_filename=test_offline_wenet_ctc.jar
kotlinc-jvm -include-runtime -d $out_filename \
test_offline_wenet_ctc.kt \
FeatureConfig.kt \
HomophoneReplacerConfig.kt \
OfflineRecognizer.kt \
OfflineStream.kt \
WaveReader.kt \
faked-asset-manager.kt
ls -lh $out_filename
java -Djava.library.path=../build/lib -jar $out_filename
}
testVersion
testOfflineWenetCtc
testOfflineNeMoCanary
testOfflineSenseVoiceWithHr
testOfflineSpeechDenoiser
... ...
package com.k2fsa.sherpa.onnx
fun main() {
val recognizer = createOfflineRecognizer()
val waveFilename = "./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav"
val objArray = WaveReader.readWaveFromFile(
filename = waveFilename,
)
val samples: FloatArray = objArray[0] as FloatArray
val sampleRate: Int = objArray[1] as Int
var stream = recognizer.createStream()
stream.acceptWaveform(samples, sampleRate=sampleRate)
recognizer.decode(stream)
var result = recognizer.getResult(stream)
println(result)
stream.release()
recognizer.release()
}
fun createOfflineRecognizer(): OfflineRecognizer {
val config = OfflineRecognizerConfig(
modelConfig = getOfflineModelConfig(type = 42)!!,
)
return OfflineRecognizer(config = config)
}
... ...
... ... @@ -124,6 +124,7 @@ The following tables list the examples in this folder.
|[./test_asr_non_streaming_moonshine.js](./test_asr_non_streaming_moonshine.js)|Non-streaming speech recognition from a file using [Moonshine](https://github.com/usefulsensors/moonshine)|
|[./test_vad_with_non_streaming_asr_moonshine.js](./test_vad_with_non_streaming_asr_moonshine.js)| Non-streaming speech recognition from a file using [Moonshine](https://github.com/usefulsensors/moonshine) + [Silero VAD](https://github.com/snakers4/silero-vad)|
|[./test_asr_non_streaming_nemo_ctc.js](./test_asr_non_streaming_nemo_ctc.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) CTC model with greedy search|
|[./test_asr_non_streaming_wenet_ctc.js](./test_asr_non_streaming_wenet_ctc.js)|Non-streaming speech recognition from a file using a [u2pp_conformer_yue](https://huggingface.co/ASLP-lab/WSYue-ASR/tree/main/u2pp_conformer_yue) CTC model with greedy search|
|[./test_asr_non_streaming_nemo_canary.js](./test_asr_non_streaming_nemo_canary.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) [Canary](https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html#sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8-english-spanish-german-french) model|
|[./test_asr_non_streaming_zipformer_ctc.js](./test_asr_non_streaming_zipformer_ctc.js)|Non-streaming speech recognition from a file using a Zipformer CTC model with greedy search|
|[./test_asr_non_streaming_nemo_parakeet_tdt_v2.js](./test_asr_non_streaming_nemo_parakeet_tdt_v2.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) [parakeet-tdt-0.6b-v2](https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-transducer/nemo-transducer-models.html#sherpa-onnx-nemo-parakeet-tdt-0-6b-v2-int8-english) model with greedy search|
... ... @@ -426,6 +427,16 @@ npm install naudiodon2
node ./test_vad_asr_non_streaming_nemo_ctc_microphone.js
```
### Non-streaming speech recognition with Wenet CTC models
```bash
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
node ./test_asr_non_streaming_wenet_ctc.js
```
### Non-streaming speech recognition with Paraformer
```bash
... ...
// Copyright (c) 2024 Xiaomi Corporation
const sherpa_onnx = require('sherpa-onnx-node');
// Please download test files from
// https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
const config = {
'featConfig': {
'sampleRate': 16000,
'featureDim': 80,
},
'modelConfig': {
'wenetCtc': {
'model':
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx',
},
'tokens':
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt',
'numThreads': 2,
'provider': 'cpu',
'debug': 1,
}
};
const waveFilename =
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav';
const recognizer = new sherpa_onnx.OfflineRecognizer(config);
console.log('Started')
let start = Date.now();
const stream = recognizer.createStream();
const wave = sherpa_onnx.readWave(waveFilename);
stream.acceptWaveform({sampleRate: wave.sampleRate, samples: wave.samples});
recognizer.decode(stream);
result = recognizer.getResult(stream)
let stop = Date.now();
console.log('Done')
const elapsed_seconds = (stop - start) / 1000;
const duration = wave.samples.length / wave.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))
console.log(waveFilename)
console.log('result\n', result)
... ...
... ... @@ -203,6 +203,22 @@ rm sherpa-onnx-zipformer-ctc-zh-int8-2025-07-03.tar.bz2
node ./test-offline-zipformer-ctc.js
```
## ./test-offline-wenet-ctc.js
[./test-offline-wenet-ctc.js](./test-offline-wenet-ctc.js) demonstrates
how to decode a file with a Wenet CTC model. In the code we use
[sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2](https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2).
You can use the following command to run it:
```bash
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
node ./test-offline-wenet-ctc.js
```
## ./test-offline-nemo-ctc.js
[./test-offline-nemo-ctc.js](./test-offline-nemo-ctc.js) demonstrates
... ...
// Copyright (c) 2025 Xiaomi Corporation (authors: Fangjun Kuang)
//
const fs = require('fs');
const {Readable} = require('stream');
const wav = require('wav');
const sherpa_onnx = require('sherpa-onnx');
function createOfflineRecognizer() {
let config = {
modelConfig: {
wenetCtc: {
model:
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx',
},
tokens:
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt',
}
};
return sherpa_onnx.createOfflineRecognizer(config);
}
const recognizer = createOfflineRecognizer();
const stream = recognizer.createStream();
const waveFilename =
'./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav';
const wave = sherpa_onnx.readWave(waveFilename);
stream.acceptWaveform(wave.sampleRate, wave.samples);
recognizer.decode(stream);
const text = recognizer.getResult(stream).text;
console.log(text);
stream.free();
recognizer.free();
... ...
... ... @@ -10,4 +10,5 @@ telespeech_ctc
moonshine
dolphin_ctc
zipformer_ctc
wenet_ctc
nemo_canary
... ...
#!/usr/bin/env bash
set -ex
SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
SHERPA_ONNX_DIR=$(cd $SCRIPT_DIR/../.. && pwd)
echo "SHERPA_ONNX_DIR: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
ls -lh lib
popd
fi
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
fpc \
-dSHERPA_ONNX_USE_SHARED_LIBS \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
./wenet_ctc.pas
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./wenet_ctc
... ...
{ Copyright (c) 2025 Xiaomi Corporation }
{
This file shows how to use a non-streaming Wenet CTC model
to decode files.
You can download the model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
}
program wenet_ctc;
{$mode objfpc}
uses
sherpa_onnx,
DateUtils,
SysUtils;
var
Wave: TSherpaOnnxWave;
WaveFilename: AnsiString;
Config: TSherpaOnnxOfflineRecognizerConfig;
Recognizer: TSherpaOnnxOfflineRecognizer;
Stream: TSherpaOnnxOfflineStream;
RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
Start: TDateTime;
Stop: TDateTime;
Elapsed: Single;
Duration: Single;
RealTimeFactor: Single;
begin
Initialize(Config);
Config.ModelConfig.WenetCtc.Model := './sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx';
Config.ModelConfig.Tokens := './sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt';
Config.ModelConfig.Provider := 'cpu';
Config.ModelConfig.NumThreads := 1;
Config.ModelConfig.Debug := False;
WaveFilename := './sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav';
Wave := SherpaOnnxReadWave(WaveFilename);
Recognizer := TSherpaOnnxOfflineRecognizer.Create(Config);
Stream := Recognizer.CreateStream();
Start := Now;
Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Stop := Now;
Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
Duration := Length(Wave.Samples) / Wave.SampleRate;
RealTimeFactor := Elapsed / Duration;
WriteLn(RecognitionResult.ToString);
WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
WriteLn(Format('Elapsed %.3f s', [Elapsed]));
WriteLn(Format('Wave duration %.3f s', [Duration]));
WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
{Free resources to avoid memory leak.
Note: You don't need to invoke them for this simple script.
However, you have to invoke them in your own large/complex project.
}
FreeAndNil(Stream);
FreeAndNil(Recognizer);
end.
... ...
... ... @@ -184,7 +184,6 @@ def get_2nd_models():
pushd $model_name
rm -rfv test_wavs
rm -fv model.onnx
rm -fv *.py
ls -lh
... ... @@ -192,6 +191,21 @@ def get_2nd_models():
popd
""",
),
Model(
model_name="sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10",
idx=42,
lang="zh_en_yue",
short_name="wenetspeech_yue_u2pconformer_ctc_2025_09_10_int8",
cmd="""
pushd $model_name
rm -rfv test_wavs
ls -lh
popd
""",
),
]
return models
... ... @@ -399,6 +413,7 @@ def get_models():
"sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17",
"sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2025-09-09",
"sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02",
"sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10",
]
for first_m in first_zh:
for second_m in second_zh:
... ... @@ -425,6 +440,10 @@ def get_models():
"sherpa-onnx-streaming-zipformer-en-20M-2023-02-17",
"sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2025-09-09",
),
(
"sherpa-onnx-streaming-zipformer-en-20M-2023-02-17",
"sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10",
),
]
models = []
for f, s in combinations:
... ...
... ... @@ -714,6 +714,22 @@ def get_models():
popd
""",
),
Model(
model_name="sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10",
idx=42,
lang="zh_en_yue",
lang2="中英粤",
short_name="wenetspeech_yue_u2pconformer_ctc_2025_09_10_int8",
cmd="""
pushd $model_name
rm -rfv test_wavs
ls -lh
popd
""",
),
]
return models
... ...
... ... @@ -29,6 +29,7 @@ namespace SherpaOnnx
Dolphin = new OfflineDolphinModelConfig();
ZipformerCtc = new OfflineZipformerCtcModelConfig();
Canary = new OfflineCanaryModelConfig();
WenetCtc = new OfflineWenetCtcModelConfig();
}
public OfflineTransducerModelConfig Transducer;
public OfflineParaformerModelConfig Paraformer;
... ... @@ -64,5 +65,6 @@ namespace SherpaOnnx
public OfflineDolphinModelConfig Dolphin;
public OfflineZipformerCtcModelConfig ZipformerCtc;
public OfflineCanaryModelConfig Canary;
public OfflineWenetCtcModelConfig WenetCtc;
}
}
... ...
/// Copyright (c) 2025 Xiaomi Corporation (authors: Fangjun Kuang)
using System.Runtime.InteropServices;
namespace SherpaOnnx
{
[StructLayout(LayoutKind.Sequential)]
public struct OfflineWenetCtcModelConfig
{
public OfflineWenetCtcModelConfig()
{
Model = "";
}
[MarshalAs(UnmanagedType.LPStr)]
public string Model;
}
}
... ...
../../../../go-api-examples/non-streaming-decode-files/run-wenet-ctc.sh
\ No newline at end of file
... ...
... ... @@ -418,6 +418,10 @@ type OfflineZipformerCtcModelConfig struct {
Model string // Path to the model, e.g., model.onnx or model.int8.onnx
}
type OfflineWenetCtcModelConfig struct {
Model string // Path to the model, e.g., model.onnx or model.int8.onnx
}
type OfflineDolphinModelConfig struct {
Model string // Path to the model, e.g., model.onnx or model.int8.onnx
}
... ... @@ -478,6 +482,7 @@ type OfflineModelConfig struct {
Dolphin OfflineDolphinModelConfig
ZipformerCtc OfflineZipformerCtcModelConfig
Canary OfflineCanaryModelConfig
WenetCtc OfflineWenetCtcModelConfig
Tokens string // Path to tokens.txt
// Number of threads to use for neural network computation
... ... @@ -579,6 +584,8 @@ func newCOfflineRecognizerConfig(config *OfflineRecognizerConfig) *C.struct_Sher
c.model_config.canary.tgt_lang = C.CString(config.ModelConfig.Canary.TgtLang)
c.model_config.canary.use_pnc = C.int(config.ModelConfig.Canary.UsePnc)
c.model_config.wenet_ctc.model = C.CString(config.ModelConfig.WenetCtc.Model)
c.model_config.tokens = C.CString(config.ModelConfig.Tokens)
c.model_config.num_threads = C.int(config.ModelConfig.NumThreads)
... ... @@ -727,6 +734,11 @@ func freeCOfflineRecognizerConfig(c *C.struct_SherpaOnnxOfflineRecognizerConfig)
c.model_config.canary.tgt_lang = nil
}
if c.model_config.wenet_ctc.model != nil {
C.free(unsafe.Pointer(c.model_config.wenet_ctc.model))
c.model_config.wenet_ctc.model = nil
}
if c.model_config.tokens != nil {
C.free(unsafe.Pointer(c.model_config.tokens))
c.model_config.tokens = nil
... ...
... ... @@ -506,6 +506,9 @@ static sherpa_onnx::OfflineRecognizerConfig GetOfflineRecognizerConfig(
recognizer_config.model_config.canary.use_pnc =
config->model_config.canary.use_pnc;
recognizer_config.model_config.wenet_ctc.model =
SHERPA_ONNX_OR(config->model_config.wenet_ctc.model, "");
recognizer_config.lm_config.model =
SHERPA_ONNX_OR(config->lm_config.model, "");
recognizer_config.lm_config.scale =
... ...
... ... @@ -476,6 +476,10 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineZipformerCtcModelConfig {
const char *model;
} SherpaOnnxOfflineZipformerCtcModelConfig;
SHERPA_ONNX_API typedef struct SherpaOnnxOfflineWenetCtcModelConfig {
const char *model;
} SherpaOnnxOfflineWenetCtcModelConfig;
SHERPA_ONNX_API typedef struct SherpaOnnxOfflineModelConfig {
SherpaOnnxOfflineTransducerModelConfig transducer;
SherpaOnnxOfflineParaformerModelConfig paraformer;
... ... @@ -501,6 +505,7 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineModelConfig {
SherpaOnnxOfflineDolphinModelConfig dolphin;
SherpaOnnxOfflineZipformerCtcModelConfig zipformer_ctc;
SherpaOnnxOfflineCanaryModelConfig canary;
SherpaOnnxOfflineWenetCtcModelConfig wenet_ctc;
} SherpaOnnxOfflineModelConfig;
SHERPA_ONNX_API typedef struct SherpaOnnxOfflineRecognizerConfig {
... ...
... ... @@ -265,6 +265,8 @@ static SherpaOnnxOfflineRecognizerConfig Convert(
c.model_config.canary.tgt_lang = config.model_config.canary.tgt_lang.c_str();
c.model_config.canary.use_pnc = config.model_config.canary.use_pnc;
c.model_config.wenet_ctc.model = config.model_config.wenet_ctc.model.c_str();
c.lm_config.model = config.lm_config.model.c_str();
c.lm_config.scale = config.lm_config.scale;
... ...
... ... @@ -264,6 +264,10 @@ struct SHERPA_ONNX_API OfflineZipformerCtcModelConfig {
std::string model;
};
struct SHERPA_ONNX_API OfflineWenetCtcModelConfig {
std::string model;
};
struct SHERPA_ONNX_API OfflineMoonshineModelConfig {
std::string preprocessor;
std::string encoder;
... ... @@ -292,6 +296,7 @@ struct SHERPA_ONNX_API OfflineModelConfig {
OfflineDolphinModelConfig dolphin;
OfflineZipformerCtcModelConfig zipformer_ctc;
OfflineCanaryModelConfig canary;
OfflineWenetCtcModelConfig wenet_ctc;
};
struct SHERPA_ONNX_API OfflineLMConfig {
... ...
... ... @@ -36,6 +36,7 @@ java_files += OfflineFireRedAsrModelConfig.java
java_files += OfflineMoonshineModelConfig.java
java_files += OfflineNemoEncDecCtcModelConfig.java
java_files += OfflineZipformerCtcModelConfig.java
java_files += OfflineWenetCtcModelConfig.java
java_files += OfflineCanaryModelConfig.java
java_files += OfflineSenseVoiceModelConfig.java
java_files += OfflineDolphinModelConfig.java
... ...
... ... @@ -12,6 +12,7 @@ public class OfflineModelConfig {
private final OfflineSenseVoiceModelConfig senseVoice;
private final OfflineDolphinModelConfig dolphin;
private final OfflineZipformerCtcModelConfig zipformerCtc;
private final OfflineWenetCtcModelConfig wenetCtc;
private final OfflineCanaryModelConfig canary;
private final String teleSpeech;
private final String tokens;
... ... @@ -32,6 +33,7 @@ public class OfflineModelConfig {
this.nemo = builder.nemo;
this.zipformerCtc = builder.zipformerCtc;
this.canary = builder.canary;
this.wenetCtc = builder.wenetCtc;
this.senseVoice = builder.senseVoice;
this.dolphin = builder.dolphin;
this.teleSpeech = builder.teleSpeech;
... ... @@ -80,6 +82,10 @@ public class OfflineModelConfig {
return zipformerCtc;
}
public OfflineWenetCtcModelConfig getWenetCtc() {
return wenetCtc;
}
public OfflineCanaryModelConfig getCanary() {
return canary;
}
... ... @@ -126,6 +132,7 @@ public class OfflineModelConfig {
private OfflineSenseVoiceModelConfig senseVoice = OfflineSenseVoiceModelConfig.builder().build();
private OfflineDolphinModelConfig dolphin = OfflineDolphinModelConfig.builder().build();
private OfflineZipformerCtcModelConfig zipformerCtc = OfflineZipformerCtcModelConfig.builder().build();
private OfflineWenetCtcModelConfig wenetCtc = OfflineWenetCtcModelConfig.builder().build();
private OfflineCanaryModelConfig canary = OfflineCanaryModelConfig.builder().build();
private String teleSpeech = "";
private String tokens = "";
... ... @@ -165,6 +172,11 @@ public class OfflineModelConfig {
return this;
}
public Builder setWenetCtc(OfflineWenetCtcModelConfig wenetCtc) {
this.wenetCtc = wenetCtc;
return this;
}
public Builder setCanary(OfflineCanaryModelConfig canary) {
this.canary = canary;
return this;
... ...
package com.k2fsa.sherpa.onnx;
public class OfflineWenetCtcModelConfig {
private final String model;
private OfflineWenetCtcModelConfig(Builder builder) {
this.model = builder.model;
}
public static Builder builder() {
return new Builder();
}
public String getModel() {
return model;
}
public static class Builder {
private String model = "";
public OfflineWenetCtcModelConfig build() {
return new OfflineWenetCtcModelConfig(this);
}
public Builder setModel(String model) {
this.model = model;
return this;
}
}
}
... ...
... ... @@ -284,6 +284,19 @@ static OfflineRecognizerConfig GetOfflineConfig(JNIEnv *env, jobject config) {
ans.model_config.zipformer_ctc.model = p;
env->ReleaseStringUTFChars(s, p);
// wenet ctc
fid = env->GetFieldID(model_config_cls, "wenetCtc",
"Lcom/k2fsa/sherpa/onnx/OfflineWenetCtcModelConfig;");
jobject wenet_ctc_config = env->GetObjectField(model_config, fid);
jclass wenet_ctc_config_cls = env->GetObjectClass(wenet_ctc_config);
fid = env->GetFieldID(wenet_ctc_config_cls, "model", "Ljava/lang/String;");
s = (jstring)env->GetObjectField(wenet_ctc_config, fid);
p = env->GetStringUTFChars(s, nullptr);
ans.model_config.wenet_ctc.model = p;
env->ReleaseStringUTFChars(s, p);
// canary
fid = env->GetFieldID(model_config_cls, "canary",
"Lcom/k2fsa/sherpa/onnx/OfflineCanaryModelConfig;");
... ...
... ... @@ -36,6 +36,10 @@ data class OfflineZipformerCtcModelConfig(
var model: String = "",
)
data class OfflineWenetCtcModelConfig(
var model: String = "",
)
data class OfflineWhisperModelConfig(
var encoder: String = "",
var decoder: String = "",
... ... @@ -80,6 +84,7 @@ data class OfflineModelConfig(
var senseVoice: OfflineSenseVoiceModelConfig = OfflineSenseVoiceModelConfig(),
var dolphin: OfflineDolphinModelConfig = OfflineDolphinModelConfig(),
var zipformerCtc: OfflineZipformerCtcModelConfig = OfflineZipformerCtcModelConfig(),
var wenetCtc: OfflineWenetCtcModelConfig = OfflineWenetCtcModelConfig(),
var canary: OfflineCanaryModelConfig = OfflineCanaryModelConfig(),
var teleSpeech: String = "",
var numThreads: Int = 1,
... ... @@ -705,6 +710,16 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? {
tokens = "$modelDir/tokens.txt",
)
}
42 -> {
val modelDir = "sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10"
return OfflineModelConfig(
wenetCtc = OfflineWenetCtcModelConfig(
model = "$modelDir/model.int8.onnx",
),
tokens = "$modelDir/tokens.txt",
)
}
}
return null
}
... ...
... ... @@ -313,6 +313,11 @@ type
function ToString: AnsiString;
end;
TSherpaOnnxOfflineWenetCtcModelConfig = record
Model: AnsiString;
function ToString: AnsiString;
end;
TSherpaOnnxOfflineWhisperModelConfig = record
Encoder: AnsiString;
Decoder: AnsiString;
... ... @@ -387,6 +392,7 @@ type
Dolphin: TSherpaOnnxOfflineDolphinModelConfig;
ZipformerCtc: TSherpaOnnxOfflineZipformerCtcModelConfig;
Canary: TSherpaOnnxOfflineCanaryModelConfig;
WenetCtc: TSherpaOnnxOfflineWenetCtcModelConfig;
class operator Initialize({$IFDEF FPC}var{$ELSE}out{$ENDIF} Dest: TSherpaOnnxOfflineModelConfig);
function ToString: AnsiString;
end;
... ... @@ -794,6 +800,9 @@ type
SherpaOnnxOfflineZipformerCtcModelConfig = record
Model: PAnsiChar;
end;
SherpaOnnxOfflineWenetCtcModelConfig = record
Model: PAnsiChar;
end;
SherpaOnnxOfflineWhisperModelConfig = record
Encoder: PAnsiChar;
Decoder: PAnsiChar;
... ... @@ -850,6 +859,7 @@ type
Dolphin: SherpaOnnxOfflineDolphinModelConfig;
ZipformerCtc: SherpaOnnxOfflineZipformerCtcModelConfig;
Canary: SherpaOnnxOfflineCanaryModelConfig;
WenetCtc: SherpaOnnxOfflineWenetCtcModelConfig;
end;
SherpaOnnxOfflineRecognizerConfig = record
... ... @@ -1658,6 +1668,12 @@ begin
[Self.Model]);
end;
function TSherpaOnnxOfflineWenetCtcModelConfig.ToString: AnsiString;
begin
Result := Format('TSherpaOnnxOfflineWenetCtcModelConfig(Model := %s)',
[Self.Model]);
end;
function TSherpaOnnxOfflineWhisperModelConfig.ToString: AnsiString;
begin
Result := Format('TSherpaOnnxOfflineWhisperModelConfig(' +
... ... @@ -1747,7 +1763,8 @@ begin
'FireRedAsr := %s, ' +
'Dolphin := %s, ' +
'ZipformerCtc := %s, ' +
'Canary := %s' +
'Canary := %s, ' +
'WenetCtc := %s' +
')',
[Self.Transducer.ToString, Self.Paraformer.ToString,
Self.NeMoCtc.ToString, Self.Whisper.ToString, Self.Tdnn.ToString,
... ... @@ -1755,7 +1772,7 @@ begin
Self.ModelType, Self.ModelingUnit, Self.BpeVocab,
Self.TeleSpeechCtc, Self.SenseVoice.ToString, Self.Moonshine.ToString,
Self.FireRedAsr.ToString, Self.Dolphin.ToString,
Self.ZipformerCtc.ToString, Self.Canary.ToString
Self.ZipformerCtc.ToString, Self.Canary.ToString, Self.WenetCtc.ToString
]);
end;
... ... @@ -1834,6 +1851,8 @@ begin
C.ModelConfig.Canary.TgtLang := PAnsiChar(Config.ModelConfig.Canary.TgtLang);
C.ModelConfig.Canary.UsePnc := Ord(Config.ModelConfig.Canary.UsePnc);
C.ModelConfig.WenetCtc.Model := PAnsiChar(Config.ModelConfig.WenetCtc.Model);
C.LMConfig.Model := PAnsiChar(Config.LMConfig.Model);
C.LMConfig.Scale := Config.LMConfig.Scale;
... ...
... ... @@ -19,6 +19,7 @@ speech-enhancement-gtcrn
decode-file-sense-voice-with-hr
test-version
zipformer-ctc-asr
wenet-ctc-asr
dolphin-ctc-asr
tts-kitten-en
compute-speaker-embeddings
... ...
... ... @@ -360,6 +360,14 @@ func sherpaOnnxOfflineZipformerCtcModelConfig(
)
}
func sherpaOnnxOfflineWenetCtcModelConfig(
model: String = ""
) -> SherpaOnnxOfflineWenetCtcModelConfig {
return SherpaOnnxOfflineWenetCtcModelConfig(
model: toCPointer(model)
)
}
func sherpaOnnxOfflineNemoEncDecCtcModelConfig(
model: String = ""
) -> SherpaOnnxOfflineNemoEncDecCtcModelConfig {
... ... @@ -482,7 +490,9 @@ func sherpaOnnxOfflineModelConfig(
dolphin: SherpaOnnxOfflineDolphinModelConfig = sherpaOnnxOfflineDolphinModelConfig(),
zipformerCtc: SherpaOnnxOfflineZipformerCtcModelConfig =
sherpaOnnxOfflineZipformerCtcModelConfig(),
canary: SherpaOnnxOfflineCanaryModelConfig = sherpaOnnxOfflineCanaryModelConfig()
canary: SherpaOnnxOfflineCanaryModelConfig = sherpaOnnxOfflineCanaryModelConfig(),
wenetCtc: SherpaOnnxOfflineWenetCtcModelConfig =
sherpaOnnxOfflineWenetCtcModelConfig()
) -> SherpaOnnxOfflineModelConfig {
return SherpaOnnxOfflineModelConfig(
transducer: transducer,
... ... @@ -503,7 +513,8 @@ func sherpaOnnxOfflineModelConfig(
fire_red_asr: fireRedAsr,
dolphin: dolphin,
zipformer_ctc: zipformerCtc,
canary: canary
canary: canary,
wenet_ctc: wenetCtc
)
}
... ...
#!/usr/bin/env bash
set -ex
if [ ! -d ../build-swift-macos ]; then
echo "Please run ../build-swift-macos.sh first!"
exit 1
fi
if [ ! -f sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
tar xvf sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
rm sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10.tar.bz2
fi
if [ ! -e ./wenet-ctc-asr ]; then
# Note: We use -lc++ to link against libc++ instead of libstdc++
swiftc \
-lc++ \
-I ../build-swift-macos/install/include \
-import-objc-header ./SherpaOnnx-Bridging-Header.h \
./wenet-ctc-asr.swift ./SherpaOnnx.swift \
-L ../build-swift-macos/install/lib/ \
-l sherpa-onnx \
-l onnxruntime \
-o wenet-ctc-asr
strip wenet-ctc-asr
else
echo "./wenet-ctc-asr exists - skip building"
fi
export DYLD_LIBRARY_PATH=$PWD/../build-swift-macos/install/lib:$DYLD_LIBRARY_PATH
./wenet-ctc-asr
... ...
import AVFoundation
extension AudioBuffer {
func array() -> [Float] {
return Array(UnsafeBufferPointer(self))
}
}
extension AVAudioPCMBuffer {
func array() -> [Float] {
return self.audioBufferList.pointee.mBuffers.array()
}
}
func run() {
let model =
"./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/model.int8.onnx"
let tokens =
"./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/tokens.txt"
let wenetCtc = sherpaOnnxOfflineWenetCtcModelConfig(
model: model
)
let modelConfig = sherpaOnnxOfflineModelConfig(
tokens: tokens,
debug: 0,
wenetCtc: wenetCtc
)
let featConfig = sherpaOnnxFeatureConfig(
sampleRate: 16000,
featureDim: 80
)
var config = sherpaOnnxOfflineRecognizerConfig(
featConfig: featConfig,
modelConfig: modelConfig
)
let recognizer = SherpaOnnxOfflineRecognizer(config: &config)
let filePath =
"./sherpa-onnx-wenetspeech-yue-u2pp-conformer-ctc-zh-en-cantonese-int8-2025-09-10/test_wavs/yue-0.wav"
let fileURL: NSURL = NSURL(fileURLWithPath: filePath)
let audioFile = try! AVAudioFile(forReading: fileURL as URL)
let audioFormat = audioFile.processingFormat
assert(audioFormat.channelCount == 1)
assert(audioFormat.commonFormat == AVAudioCommonFormat.pcmFormatFloat32)
let audioFrameCount = UInt32(audioFile.length)
let audioFileBuffer = AVAudioPCMBuffer(pcmFormat: audioFormat, frameCapacity: audioFrameCount)
try! audioFile.read(into: audioFileBuffer!)
let array: [Float]! = audioFileBuffer?.array()
let result = recognizer.decode(samples: array, sampleRate: Int(audioFormat.sampleRate))
print("\nresult is:\n\(result.text)")
if result.timestamps.count != 0 {
print("\ntimestamps is:\n\(result.timestamps)")
}
}
@main
struct App {
static func main() {
run()
}
}
... ...
... ... @@ -51,6 +51,10 @@ function freeConfig(config, Module) {
freeConfig(config.zipformerCtc, Module)
}
if ('wenetCtc' in config) {
freeConfig(config.wenetCtc, Module)
}
if ('moonshine' in config) {
freeConfig(config.moonshine, Module)
}
... ... @@ -733,6 +737,23 @@ function initSherpaOnnxOfflineZipformerCtcModelConfig(config, Module) {
}
}
function initSherpaOnnxOfflineWenetCtcModelConfig(config, Module) {
const n = Module.lengthBytesUTF8(config.model || '') + 1;
const buffer = Module._malloc(n);
const len = 1 * 4; // 1 pointer
const ptr = Module._malloc(len);
Module.stringToUTF8(config.model || '', buffer, n);
Module.setValue(ptr, buffer, 'i8*');
return {
buffer: buffer, ptr: ptr, len: len,
}
}
function initSherpaOnnxOfflineWhisperModelConfig(config, Module) {
const encoderLen = Module.lengthBytesUTF8(config.encoder || '') + 1;
const decoderLen = Module.lengthBytesUTF8(config.decoder || '') + 1;
... ... @@ -997,6 +1018,12 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
};
}
if (!('wenetCtc' in config)) {
config.wenetCtc = {
model: '',
};
}
if (!('whisper' in config)) {
config.whisper = {
encoder: '',
... ... @@ -1078,9 +1105,12 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
const canary = initSherpaOnnxOfflineCanaryModelConfig(config.canary, Module);
const wenetCtc =
initSherpaOnnxOfflineWenetCtcModelConfig(config.wenetCtc, Module);
const len = transducer.len + paraformer.len + nemoCtc.len + whisper.len +
tdnn.len + 8 * 4 + senseVoice.len + moonshine.len + fireRedAsr.len +
dolphin.len + zipformerCtc.len + canary.len;
dolphin.len + zipformerCtc.len + canary.len + wenetCtc.len;
const ptr = Module._malloc(len);
... ... @@ -1188,11 +1218,15 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
Module._CopyHeap(canary.ptr, canary.len, ptr + offset);
offset += canary.len;
Module._CopyHeap(wenetCtc.ptr, wenetCtc.len, ptr + offset);
offset += wenetCtc.len;
return {
buffer: buffer, ptr: ptr, len: len, transducer: transducer,
paraformer: paraformer, nemoCtc: nemoCtc, whisper: whisper, tdnn: tdnn,
senseVoice: senseVoice, moonshine: moonshine, fireRedAsr: fireRedAsr,
dolphin: dolphin, zipformerCtc: zipformerCtc, canary: canary,
wenetCtc: wenetCtc,
}
}
... ...
... ... @@ -14,6 +14,7 @@ static_assert(sizeof(SherpaOnnxOfflineTransducerModelConfig) == 3 * 4, "");
static_assert(sizeof(SherpaOnnxOfflineParaformerModelConfig) == 4, "");
static_assert(sizeof(SherpaOnnxOfflineZipformerCtcModelConfig) == 4, "");
static_assert(sizeof(SherpaOnnxOfflineWenetCtcModelConfig) == 4, "");
static_assert(sizeof(SherpaOnnxOfflineDolphinModelConfig) == 4, "");
static_assert(sizeof(SherpaOnnxOfflineNemoEncDecCtcModelConfig) == 4, "");
static_assert(sizeof(SherpaOnnxOfflineWhisperModelConfig) == 5 * 4, "");
... ... @@ -35,7 +36,8 @@ static_assert(sizeof(SherpaOnnxOfflineModelConfig) ==
sizeof(SherpaOnnxOfflineFireRedAsrModelConfig) +
sizeof(SherpaOnnxOfflineDolphinModelConfig) +
sizeof(SherpaOnnxOfflineZipformerCtcModelConfig) +
sizeof(SherpaOnnxOfflineCanaryModelConfig),
sizeof(SherpaOnnxOfflineCanaryModelConfig) +
sizeof(SherpaOnnxOfflineWenetCtcModelConfig),
"");
static_assert(sizeof(SherpaOnnxFeatureConfig) == 2 * 4, "");
... ... @@ -83,6 +85,7 @@ void PrintOfflineRecognizerConfig(SherpaOnnxOfflineRecognizerConfig *config) {
auto dolphin = &model_config->dolphin;
auto zipformer_ctc = &model_config->zipformer_ctc;
auto canary = &model_config->canary;
auto wenet_ctc = &model_config->wenet_ctc;
fprintf(stdout, "----------offline transducer model config----------\n");
fprintf(stdout, "encoder: %s\n", transducer->encoder);
... ... @@ -133,6 +136,9 @@ void PrintOfflineRecognizerConfig(SherpaOnnxOfflineRecognizerConfig *config) {
fprintf(stdout, "tgt_lang: %s\n", canary->tgt_lang);
fprintf(stdout, "use_pnc: %d\n", canary->use_pnc);
fprintf(stdout, "----------offline wenet ctc model config----------\n");
fprintf(stdout, "model: %s\n", wenet_ctc->model);
fprintf(stdout, "tokens: %s\n", model_config->tokens);
fprintf(stdout, "num_threads: %d\n", model_config->num_threads);
fprintf(stdout, "provider: %s\n", model_config->provider);
... ...