Add dart API for SenseVoice (#1159)

.github/scripts/test-dart.sh

@@ -6,6 +6,10 @@ cd dart-api-examples
 
 pushd non-streaming-asr
 
+echo '----------SenseVoice----------'
+./run-sense-voice.sh
+rm -rf sherpa-onnx-*
+
 echo '----------NeMo transducer----------'
 ./run-nemo-transducer.sh
 rm -rf sherpa-onnx-*

dart-api-examples/non-streaming-asr/README.md

@@ -11,4 +11,5 @@ This folder contains examples for non-streaming ASR with Dart API.
 |[./bin/whisper.dart](./bin/whisper.dart)| Use whisper for speech recognition. See [./run-whisper.sh](./run-whisper.sh)|
 |[./bin/zipformer-transducer.dart](./bin/zipformer-transducer.dart)| Use a zipformer transducer for speech recognition. See [./run-zipformer-transducer.sh](./run-zipformer-transducer.sh)|
 |[./bin/vad-with-paraformer.dart](./bin/vad-with-paraformer.dart)| Use a [silero-vad](https://github.com/snakers4/silero-vad) with paraformer for speech recognition. See [./run-vad-with-paraformer.sh](./run-vad-with-paraformer.sh)|
+|[./bin/sense-voice.dart](./bin/sense-voice.dart)| Use a SenseVoice CTC model for speech recognition. See [./run-sense-voice.sh](./run-sense-voice.sh)|
 

dart-api-examples/non-streaming-asr/bin/sense-voice.dart

@@ -0,0 +1,61 @@
+// Copyright (c)  2024  Xiaomi Corporation
+import 'dart:io';
+import 'dart:typed_data';
+
+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 paraformer model')
+    ..addOption('tokens', help: 'Path to tokens.txt')
+    ..addOption('language',
+        help: 'auto, zh, en, ja, ko, yue, or leave it empty to use auto',
+        defaultsTo: '')
+    ..addOption('use-itn',
+        help: 'true to use inverse text normalization', defaultsTo: 'false')
+    ..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 language = res['language'] as String;
+  final useItn = (res['use-itn'] as String).toLowerCase() == 'true';
+
+  final senseVoice = sherpa_onnx.OfflineSenseVoiceModelConfig(
+      model: model, language: language, useInverseTextNormalization: useItn);
+
+  final modelConfig = sherpa_onnx.OfflineModelConfig(
+    senseVoice: senseVoice,
+    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();
+}

dart-api-examples/non-streaming-asr/pubspec.yaml

@@ -10,7 +10,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/non-streaming-asr/run-sense-voice.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+set -ex
+
+dart pub get
+
+if [ ! -f ./sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/tokens.txt ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
+  tar xvf sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
+  rm sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
+fi
+
+dart run \
+  ./bin/sense-voice.dart \
+  --model ./sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/model.int8.onnx \
+  --tokens ./sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/tokens.txt \
+  --use-itn true \
+  --input-wav ./sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/test_wavs/zh.wav

dart-api-examples/streaming-asr/pubspec.yaml

@@ -11,7 +11,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/tts/pubspec.yaml

@@ -8,7 +8,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/vad/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ^3.4.0
 
 dependencies:
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   path: ^1.9.0
   args: ^2.5.0
 

flutter-examples/streaming_asr/pubspec.yaml

@@ -5,7 +5,7 @@ description: >
 
 publish_to: 'none'
 
-version: 1.10.16
+version: 1.10.17
 
 topics:
   - speech-recognition
@@ -30,7 +30,7 @@ dependencies:
   record: ^5.1.0
   url_launcher: ^6.2.6
 
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   # sherpa_onnx:
     # path: ../../flutter/sherpa_onnx
 

flutter-examples/tts/pubspec.yaml

@@ -5,7 +5,7 @@ description: >
 
 publish_to: 'none' # Remove this line if you wish to publish to pub.dev
 
-version: 1.10.16
+version: 1.10.17
 
 environment:
   sdk: '>=3.4.0 <4.0.0'
@@ -17,7 +17,7 @@ dependencies:
   cupertino_icons: ^1.0.6
   path_provider: ^2.1.3
   path: ^1.9.0
-  sherpa_onnx: ^1.10.16
+  sherpa_onnx: ^1.10.17
   url_launcher: ^6.2.6
   audioplayers: ^5.0.0
 

flutter/sherpa_onnx/lib/src/offline_recognizer.dart

@@ -79,6 +79,23 @@ class OfflineTdnnModelConfig {
   final String model;
 }
 
+class OfflineSenseVoiceModelConfig {
+  const OfflineSenseVoiceModelConfig({
+    this.model = '',
+    this.language = '',
+    this.useInverseTextNormalization = false,
+  });
+
+  @override
+  String toString() {
+    return 'OfflineSenseVoiceModelConfig(model: $model, language: $language, useInverseTextNormalization: $useInverseTextNormalization)';
+  }
+
+  final String model;
+  final String language;
+  final bool useInverseTextNormalization;
+}
+
 class OfflineLMConfig {
   const OfflineLMConfig({this.model = '', this.scale = 1.0});
 
@@ -98,6 +115,7 @@ class OfflineModelConfig {
     this.nemoCtc = const OfflineNemoEncDecCtcModelConfig(),
     this.whisper = const OfflineWhisperModelConfig(),
     this.tdnn = const OfflineTdnnModelConfig(),
+    this.senseVoice = const OfflineSenseVoiceModelConfig(),
     required this.tokens,
     this.numThreads = 1,
     this.debug = true,
@@ -110,7 +128,7 @@ class OfflineModelConfig {
 
   @override
   String toString() {
-    return 'OfflineModelConfig(transducer: $transducer, paraformer: $paraformer, nemoCtc: $nemoCtc, whisper: $whisper, tdnn: $tdnn, 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, tokens: $tokens, numThreads: $numThreads, debug: $debug, provider: $provider, modelType: $modelType, modelingUnit: $modelingUnit, bpeVocab: $bpeVocab, telespeechCtc: $telespeechCtc)';
   }
 
   final OfflineTransducerModelConfig transducer;
@@ -118,6 +136,7 @@ class OfflineModelConfig {
   final OfflineNemoEncDecCtcModelConfig nemoCtc;
   final OfflineWhisperModelConfig whisper;
   final OfflineTdnnModelConfig tdnn;
+  final OfflineSenseVoiceModelConfig senseVoice;
 
   final String tokens;
   final int numThreads;
@@ -219,6 +238,14 @@ class OfflineRecognizer {
 
     c.ref.model.tdnn.model = config.model.tdnn.model.toNativeUtf8();
 
+    c.ref.model.senseVoice.model = config.model.senseVoice.model.toNativeUtf8();
+
+    c.ref.model.senseVoice.language =
+        config.model.senseVoice.language.toNativeUtf8();
+
+    c.ref.model.senseVoice.useInverseTextNormalization =
+        config.model.senseVoice.useInverseTextNormalization ? 1 : 0;
+
     c.ref.model.tokens = config.model.tokens.toNativeUtf8();
 
     c.ref.model.numThreads = config.model.numThreads;
@@ -254,6 +281,8 @@ 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.senseVoice.language);
+    calloc.free(c.ref.model.senseVoice.model);
     calloc.free(c.ref.model.tdnn.model);
     calloc.free(c.ref.model.whisper.task);
     calloc.free(c.ref.model.whisper.language);

flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart

@@ -87,6 +87,14 @@ final class SherpaOnnxOfflineTdnnModelConfig extends Struct {
   external Pointer<Utf8> model;
 }
 
+final class SherpaOnnxOfflineSenseVoiceModelConfig extends Struct {
+  external Pointer<Utf8> model;
+  external Pointer<Utf8> language;
+
+  @Int32()
+  external int useInverseTextNormalization;
+}
+
 final class SherpaOnnxOfflineLMConfig extends Struct {
   external Pointer<Utf8> model;
 
@@ -115,6 +123,8 @@ final class SherpaOnnxOfflineModelConfig extends Struct {
   external Pointer<Utf8> modelingUnit;
   external Pointer<Utf8> bpeVocab;
   external Pointer<Utf8> telespeechCtc;
+
+  external SherpaOnnxOfflineSenseVoiceModelConfig senseVoice;
 }
 
 final class SherpaOnnxOfflineRecognizerConfig extends Struct {

flutter/sherpa_onnx/pubspec.yaml

@@ -17,7 +17,7 @@ topics:
   - voice-activity-detection
 
 # remember to change the version in ../sherpa_onnx_macos/macos/sherpa_onnx_macos.podspec
-version: 1.10.16
+version: 1.10.17
 
 homepage: https://github.com/k2-fsa/sherpa-onnx
 
@@ -30,19 +30,19 @@ dependencies:
   flutter:
     sdk: flutter
 
-  sherpa_onnx_android: ^1.10.16
+  sherpa_onnx_android: ^1.10.17
     # path: ../sherpa_onnx_android
 
-  sherpa_onnx_macos: ^1.10.16
+  sherpa_onnx_macos: ^1.10.17
     # path: ../sherpa_onnx_macos
 
-  sherpa_onnx_linux: ^1.10.16
+  sherpa_onnx_linux: ^1.10.17
     # path: ../sherpa_onnx_linux
     #
-  sherpa_onnx_windows: ^1.10.16
+  sherpa_onnx_windows: ^1.10.17
     # path: ../sherpa_onnx_windows
 
-  sherpa_onnx_ios: ^1.10.16
+  sherpa_onnx_ios: ^1.10.17
   # sherpa_onnx_ios:
     # path: ../sherpa_onnx_ios
 

flutter/sherpa_onnx_ios/ios/sherpa_onnx_ios.podspec

@@ -7,7 +7,7 @@
 # https://groups.google.com/g/dart-ffi/c/nUATMBy7r0c
 Pod::Spec.new do |s|
   s.name             = 'sherpa_onnx_ios'
-  s.version          = '1.10.16'
+  s.version          = '1.10.17'
   s.summary          = 'A new Flutter FFI plugin project.'
   s.description      = <<-DESC
 A new Flutter FFI plugin project.

flutter/sherpa_onnx_macos/macos/sherpa_onnx_macos.podspec

@@ -4,7 +4,7 @@
 #
 Pod::Spec.new do |s|
   s.name             = 'sherpa_onnx_macos'
-  s.version          = '1.10.16'
+  s.version          = '1.10.17'
   s.summary          = 'sherpa-onnx Flutter FFI plugin project.'
   s.description      = <<-DESC
 sherpa-onnx Flutter FFI plugin project.

scripts/dart/sherpa-onnx-pubspec.yaml

@@ -17,7 +17,7 @@ topics:
   - voice-activity-detection
 
 # remember to change the version in ../sherpa_onnx_macos/macos/sherpa_onnx.podspec
-version: 1.10.16
+version: 1.10.17
 
 homepage: https://github.com/k2-fsa/sherpa-onnx
 

sherpa-onnx/csrc/provider-config.cc

@@ -13,14 +13,15 @@ namespace sherpa_onnx {
 
 void CudaConfig::Register(ParseOptions *po) {
   po->Register("cuda-cudnn-conv-algo-search", &cudnn_conv_algo_search,
-          "CuDNN convolution algrorithm search");
+               "CuDNN convolution algrorithm search");
 }
 
 bool CudaConfig::Validate() const {
   if (cudnn_conv_algo_search < 1 || cudnn_conv_algo_search > 3) {
-    SHERPA_ONNX_LOGE("cudnn_conv_algo_search: '%d' is not a valid option."
+    SHERPA_ONNX_LOGE(
-                     "Options : [1,3]. Check OnnxRT docs",
+        "cudnn_conv_algo_search: '%d' is not a valid option."
-                    cudnn_conv_algo_search);
+        "Options : [1,3]. Check OnnxRT docs",
+        cudnn_conv_algo_search);
     return false;
   }
   return true;
@@ -37,41 +38,41 @@ std::string CudaConfig::ToString() const {
 
 void TensorrtConfig::Register(ParseOptions *po) {
   po->Register("trt-max-workspace-size", &trt_max_workspace_size,
-              "Set TensorRT EP GPU memory usage limit.");
+               "Set TensorRT EP GPU memory usage limit.");
   po->Register("trt-max-partition-iterations", &trt_max_partition_iterations,
-              "Limit partitioning iterations for model conversion.");
+               "Limit partitioning iterations for model conversion.");
   po->Register("trt-min-subgraph-size", &trt_min_subgraph_size,
-              "Set minimum size for subgraphs in partitioning.");
+               "Set minimum size for subgraphs in partitioning.");
   po->Register("trt-fp16-enable", &trt_fp16_enable,
-              "Enable FP16 precision for faster performance.");
+               "Enable FP16 precision for faster performance.");
   po->Register("trt-detailed-build-log", &trt_detailed_build_log,
-              "Enable detailed logging of build steps.");
+               "Enable detailed logging of build steps.");
   po->Register("trt-engine-cache-enable", &trt_engine_cache_enable,
-              "Enable caching of TensorRT engines.");
+               "Enable caching of TensorRT engines.");
   po->Register("trt-timing-cache-enable", &trt_timing_cache_enable,
-              "Enable use of timing cache to speed up builds.");
+               "Enable use of timing cache to speed up builds.");
   po->Register("trt-engine-cache-path", &trt_engine_cache_path,
-              "Set path to store cached TensorRT engines.");
+               "Set path to store cached TensorRT engines.");
   po->Register("trt-timing-cache-path", &trt_timing_cache_path,
-              "Set path for storing timing cache.");
+               "Set path for storing timing cache.");
   po->Register("trt-dump-subgraphs", &trt_dump_subgraphs,
-              "Dump optimized subgraphs for debugging.");
+               "Dump optimized subgraphs for debugging.");
 }
 
 bool TensorrtConfig::Validate() const {
   if (trt_max_workspace_size < 0) {
-    SHERPA_ONNX_LOGE("trt_max_workspace_size: %lld is not valid.",
+    SHERPA_ONNX_LOGE("trt_max_workspace_size: %ld is not valid.",
-        trt_max_workspace_size);
+                     trt_max_workspace_size);
     return false;
   }
   if (trt_max_partition_iterations < 0) {
     SHERPA_ONNX_LOGE("trt_max_partition_iterations: %d is not valid.",
-        trt_max_partition_iterations);
+                     trt_max_partition_iterations);
     return false;
   }
   if (trt_min_subgraph_size < 0) {
     SHERPA_ONNX_LOGE("trt_min_subgraph_size: %d is not valid.",
-        trt_min_subgraph_size);
+                     trt_min_subgraph_size);
     return false;
   }
 
@@ -83,23 +84,19 @@ std::string TensorrtConfig::ToString() const {
 
   os << "TensorrtConfig(";
   os << "trt_max_workspace_size=" << trt_max_workspace_size << ", ";
-  os << "trt_max_partition_iterations="
+  os << "trt_max_partition_iterations=" << trt_max_partition_iterations << ", ";
-      << trt_max_partition_iterations << ", ";
   os << "trt_min_subgraph_size=" << trt_min_subgraph_size << ", ";
-  os << "trt_fp16_enable=\""
+  os << "trt_fp16_enable=\"" << (trt_fp16_enable ? "True" : "False") << "\", ";
-      << (trt_fp16_enable? "True" : "False") << "\", ";
   os << "trt_detailed_build_log=\""
-      << (trt_detailed_build_log? "True" : "False") << "\", ";
+     << (trt_detailed_build_log ? "True" : "False") << "\", ";
   os << "trt_engine_cache_enable=\""
-      << (trt_engine_cache_enable? "True" : "False") << "\", ";
+     << (trt_engine_cache_enable ? "True" : "False") << "\", ";
-  os << "trt_engine_cache_path=\""
+  os << "trt_engine_cache_path=\"" << trt_engine_cache_path.c_str() << "\", ";
-      << trt_engine_cache_path.c_str() << "\", ";
   os << "trt_timing_cache_enable=\""
-      << (trt_timing_cache_enable? "True" : "False") << "\", ";
+     << (trt_timing_cache_enable ? "True" : "False") << "\", ";
-  os << "trt_timing_cache_path=\""
+  os << "trt_timing_cache_path=\"" << trt_timing_cache_path.c_str() << "\",";
-      << trt_timing_cache_path.c_str() << "\",";
+  os << "trt_dump_subgraphs=\"" << (trt_dump_subgraphs ? "True" : "False")
-  os << "trt_dump_subgraphs=\""
+     << "\" )";
-      << (trt_dump_subgraphs? "True" : "False") << "\" )";
   return os.str();
 }