Add Dart API for homophone replacer (#2167)

@@ -4,6 +4,58 @@ set -ex
 
 cd dart-api-examples
 
+pushd non-streaming-asr
+
+echo '----------SenseVoice----------'
+./run-sense-voice-with-hr.sh
+./run-sense-voice.sh
+rm -rf sherpa-onnx-*
+
+echo '----------FireRedAsr----------'
+./run-fire-red-asr.sh
+rm -rf sherpa-onnx-fire-red-asr-*
+
+echo '----------NeMo transducer----------'
+./run-nemo-transducer.sh
+rm -rf sherpa-onnx-*
+
+echo '----------Dolphin CTC----------'
+./run-dolphin-ctc.sh
+rm -rf sherpa-onnx-*
+
+echo '----------NeMo CTC----------'
+./run-nemo-ctc.sh
+rm -rf sherpa-onnx-*
+
+echo '----------TeleSpeech CTC----------'
+./run-telespeech-ctc.sh
+rm -rf sherpa-onnx-*
+
+echo '----------moonshine----------'
+./run-moonshine.sh
+rm -rf sherpa-onnx-*
+
+echo '----------whisper----------'
+./run-whisper.sh
+rm -rf sherpa-onnx-*
+
+echo '----------zipformer transducer----------'
+./run-zipformer-transducer.sh
+rm -rf sherpa-onnx-*
+
+echo '----------paraformer itn----------'
+./run-paraformer-itn.sh
+
+echo '----------paraformer----------'
+./run-paraformer.sh
+rm -rf sherpa-onnx-*
+
+echo '----------VAD with paraformer----------'
+./run-vad-with-paraformer.sh
+rm -rf sherpa-onnx-*
+
+popd # non-streaming-asr
+
 pushd speech-enhancement-gtcrn
 echo "speech enhancement with gtcrn models"
 ./run.sh
@@ -101,57 +153,6 @@ pushd keyword-spotter
 ./run-zh.sh
 popd
 
-pushd non-streaming-asr
-
-echo '----------FireRedAsr----------'
-./run-fire-red-asr.sh
-rm -rf sherpa-onnx-fire-red-asr-*
-
-echo '----------SenseVoice----------'
-./run-sense-voice.sh
-rm -rf sherpa-onnx-*
-
-echo '----------NeMo transducer----------'
-./run-nemo-transducer.sh
-rm -rf sherpa-onnx-*
-
-echo '----------Dolphin CTC----------'
-./run-dolphin-ctc.sh
-rm -rf sherpa-onnx-*
-
-echo '----------NeMo CTC----------'
-./run-nemo-ctc.sh
-rm -rf sherpa-onnx-*
-
-echo '----------TeleSpeech CTC----------'
-./run-telespeech-ctc.sh
-rm -rf sherpa-onnx-*
-
-echo '----------moonshine----------'
-./run-moonshine.sh
-rm -rf sherpa-onnx-*
-
-echo '----------whisper----------'
-./run-whisper.sh
-rm -rf sherpa-onnx-*
-
-echo '----------zipformer transducer----------'
-./run-zipformer-transducer.sh
-rm -rf sherpa-onnx-*
-
-echo '----------paraformer itn----------'
-./run-paraformer-itn.sh
-
-echo '----------paraformer----------'
-./run-paraformer.sh
-rm -rf sherpa-onnx-*
-
-echo '----------VAD with paraformer----------'
-./run-vad-with-paraformer.sh
-rm -rf sherpa-onnx-*
-
-popd # non-streaming-asr
-
 pushd streaming-asr
 
 echo '----------streaming zipformer ctc HLG----------'

+// 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 SenseVoice 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')
+    ..addOption('hr-dict-dir',
+        help: 'Path to jieba dict for homophone replacer')
+    ..addOption('hr-lexicon',
+        help: 'Path to lexicon.txt for homophone replacer')
+    ..addOption('hr-rule-fsts',
+        help: 'Path to replace.fst for homophone replacer');
+
+  final res = parser.parse(arguments);
+  if (res['model'] == null ||
+      res['tokens'] == null ||
+      res['hr-dict-dir'] == null ||
+      res['hr-lexicon'] == null ||
+      res['hr-rule-fsts'] == 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 hrDictDir = res['hr-dict-dir'] as String;
+  final hrLexicon = res['hr-lexicon'] as String;
+  final hrRuleFsts = res['hr-rule-fsts'] as String;
+
+  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 hr = sherpa_onnx.HomophoneReplacerConfig(
+      dictDir: hrDictDir, lexicon: hrLexicon, ruleFsts: hrRuleFsts);
+
+  final config =
+      sherpa_onnx.OfflineRecognizerConfig(model: modelConfig, hr: hr);
+
+  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-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
+
+if [ ! -d dict ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/hr-files/dict.tar.bz2
+  tar xf dict.tar.bz2
+  rm dict.tar.bz2
+
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/hr-files/replace.fst
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/hr-files/test-hr.wav
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/hr-files/lexicon.txt
+fi
+
+dart run \
+  ./bin/sense-voice-with-hr.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 \
+  --hr-dict-dir ./dict \
+  --hr-lexicon ./lexicon.txt \
+  --hr-rule-fsts ./replace.fst \
+  --input-wav ./test-hr.wav

@@ -4,6 +4,7 @@ import 'dart:ffi';
 
 export 'src/audio_tagging.dart';
 export 'src/feature_config.dart';
+export 'src/homophone_replacer_config.dart';
 export 'src/keyword_spotter.dart';
 export 'src/offline_punctuation.dart';
 export 'src/offline_recognizer.dart';

+// Copyright (c)  2025  Xiaomi Corporation
+
+class HomophoneReplacerConfig {
+  const HomophoneReplacerConfig(
+      {this.dictDir = '', this.lexicon = '', this.ruleFsts = ''});
+
+  factory HomophoneReplacerConfig.fromJson(Map<String, dynamic> json) {
+    return HomophoneReplacerConfig(
+      dictDir: json['dictDir'] as String? ?? '',
+      lexicon: json['lexicon'] as String? ?? '',
+      ruleFsts: json['ruleFsts'] as String? ?? '',
+    );
+  }
+
+  @override
+  String toString() {
+    return 'HomophoneReplacerConfig(dictDir: $dictDir, lexicon: $lexicon, ruleFsts: $ruleFsts)';
+  }
+
+  Map<String, dynamic> toJson() => {
+        'dictDir': dictDir,
+        'lexicon': lexicon,
+        'ruleFsts': ruleFsts,
+      };
+
+  final String dictDir;
+  final String lexicon;
+  final String ruleFsts;
+}

@@ -5,6 +5,7 @@ import 'dart:ffi';
 import 'package:ffi/ffi.dart';
 
 import './feature_config.dart';
+import './homophone_replacer_config.dart';
 import './offline_stream.dart';
 import './sherpa_onnx_bindings.dart';
 import './utils.dart';
@@ -403,6 +404,7 @@ class OfflineRecognizerConfig {
     this.ruleFsts = '',
     this.ruleFars = '',
     this.blankPenalty = 0.0,
+    this.hr = const HomophoneReplacerConfig(),
   });
 
   factory OfflineRecognizerConfig.fromJson(Map<String, dynamic> json) {
@@ -421,12 +423,13 @@ class OfflineRecognizerConfig {
       ruleFsts: json['ruleFsts'] as String? ?? '',
       ruleFars: json['ruleFars'] as String? ?? '',
       blankPenalty: (json['blankPenalty'] as num?)?.toDouble() ?? 0.0,
+      hr: HomophoneReplacerConfig.fromJson(json['hr'] as Map<String, dynamic>),
     );
   }
 
   @override
   String toString() {
-    return 'OfflineRecognizerConfig(feat: $feat, model: $model, lm: $lm, decodingMethod: $decodingMethod, maxActivePaths: $maxActivePaths, hotwordsFile: $hotwordsFile, hotwordsScore: $hotwordsScore, ruleFsts: $ruleFsts, ruleFars: $ruleFars, blankPenalty: $blankPenalty)';
+    return 'OfflineRecognizerConfig(feat: $feat, model: $model, lm: $lm, decodingMethod: $decodingMethod, maxActivePaths: $maxActivePaths, hotwordsFile: $hotwordsFile, hotwordsScore: $hotwordsScore, ruleFsts: $ruleFsts, ruleFars: $ruleFars, blankPenalty: $blankPenalty, hr: $hr)';
   }
 
   Map<String, dynamic> toJson() => {
@@ -440,6 +443,7 @@ class OfflineRecognizerConfig {
         'ruleFsts': ruleFsts,
         'ruleFars': ruleFars,
         'blankPenalty': blankPenalty,
+        'hr': hr.toJson(),
       };
 
   final FeatureConfig feat;
@@ -457,6 +461,7 @@ class OfflineRecognizerConfig {
   final String ruleFars;
 
   final double blankPenalty;
+  final HomophoneReplacerConfig hr;
 }
 
 class OfflineRecognizerResult {
@@ -598,8 +603,15 @@ class OfflineRecognizer {
 
     c.ref.blankPenalty = config.blankPenalty;
 
+    c.ref.hr.dictDir = config.hr.dictDir.toNativeUtf8();
+    c.ref.hr.lexicon = config.hr.lexicon.toNativeUtf8();
+    c.ref.hr.ruleFsts = config.hr.ruleFsts.toNativeUtf8();
+
     final ptr = SherpaOnnxBindings.createOfflineRecognizer?.call(c) ?? nullptr;
 
+    calloc.free(c.ref.hr.dictDir);
+    calloc.free(c.ref.hr.lexicon);
+    calloc.free(c.ref.hr.ruleFsts);
     calloc.free(c.ref.ruleFars);
     calloc.free(c.ref.ruleFsts);
     calloc.free(c.ref.hotwordsFile);

@@ -5,6 +5,7 @@ import 'dart:ffi';
 import 'package:ffi/ffi.dart';
 
 import './feature_config.dart';
+import './homophone_replacer_config.dart';
 import './online_stream.dart';
 import './sherpa_onnx_bindings.dart';
 import './utils.dart';
@@ -194,6 +195,7 @@ class OnlineRecognizerConfig {
     this.ruleFsts = '',
     this.ruleFars = '',
     this.blankPenalty = 0.0,
+    this.hr = const HomophoneReplacerConfig(),
   });
 
   factory OnlineRecognizerConfig.fromJson(Map<String, dynamic> json) {
@@ -217,12 +219,14 @@ class OnlineRecognizerConfig {
       ruleFsts: json['ruleFsts'] as String? ?? '',
       ruleFars: json['ruleFars'] as String? ?? '',
       blankPenalty: (json['blankPenalty'] as num?)?.toDouble() ?? 0.0,
+      hr: HomophoneReplacerConfig.fromJson(
+          json['hr'] as Map<String, dynamic>? ?? const {}),
     );
   }
 
   @override
   String toString() {
-    return 'OnlineRecognizerConfig(feat: $feat, model: $model, decodingMethod: $decodingMethod, maxActivePaths: $maxActivePaths, enableEndpoint: $enableEndpoint, rule1MinTrailingSilence: $rule1MinTrailingSilence, rule2MinTrailingSilence: $rule2MinTrailingSilence, rule3MinUtteranceLength: $rule3MinUtteranceLength, hotwordsFile: $hotwordsFile, hotwordsScore: $hotwordsScore, ctcFstDecoderConfig: $ctcFstDecoderConfig, ruleFsts: $ruleFsts, ruleFars: $ruleFars, blankPenalty: $blankPenalty)';
+    return 'OnlineRecognizerConfig(feat: $feat, model: $model, decodingMethod: $decodingMethod, maxActivePaths: $maxActivePaths, enableEndpoint: $enableEndpoint, rule1MinTrailingSilence: $rule1MinTrailingSilence, rule2MinTrailingSilence: $rule2MinTrailingSilence, rule3MinUtteranceLength: $rule3MinUtteranceLength, hotwordsFile: $hotwordsFile, hotwordsScore: $hotwordsScore, ctcFstDecoderConfig: $ctcFstDecoderConfig, ruleFsts: $ruleFsts, ruleFars: $ruleFars, blankPenalty: $blankPenalty, hr: $hr)';
   }
 
   Map<String, dynamic> toJson() => {
@@ -240,6 +244,7 @@ class OnlineRecognizerConfig {
         'ruleFsts': ruleFsts,
         'ruleFars': ruleFars,
         'blankPenalty': blankPenalty,
+        'hr': hr.toJson(),
       };
 
   final FeatureConfig feat;
@@ -265,6 +270,7 @@ class OnlineRecognizerConfig {
   final String ruleFars;
 
   final double blankPenalty;
+  final HomophoneReplacerConfig hr;
 }
 
 class OnlineRecognizerResult {
@@ -352,8 +358,15 @@ class OnlineRecognizer {
 
     c.ref.blankPenalty = config.blankPenalty;
 
+    c.ref.hr.dictDir = config.hr.dictDir.toNativeUtf8();
+    c.ref.hr.lexicon = config.hr.lexicon.toNativeUtf8();
+    c.ref.hr.ruleFsts = config.hr.ruleFsts.toNativeUtf8();
+
     final ptr = SherpaOnnxBindings.createOnlineRecognizer?.call(c) ?? nullptr;
 
+    calloc.free(c.ref.hr.dictDir);
+    calloc.free(c.ref.hr.lexicon);
+    calloc.free(c.ref.hr.ruleFsts);
     calloc.free(c.ref.ruleFars);
     calloc.free(c.ref.ruleFsts);
     calloc.free(c.ref.ctcFstDecoderConfig.graph);

@@ -353,6 +353,7 @@ final class SherpaOnnxOfflineRecognizerConfig extends Struct {
 
   @Float()
   external double blankPenalty;
+  external SherpaOnnxHomophoneReplacerConfig hr;
 }
 
 final class SherpaOnnxOnlineTransducerModelConfig extends Struct {
@@ -404,6 +405,12 @@ final class SherpaOnnxOnlineCtcFstDecoderConfig extends Struct {
   external int maxActive;
 }
 
+final class SherpaOnnxHomophoneReplacerConfig extends Struct {
+  external Pointer<Utf8> dictDir;
+  external Pointer<Utf8> lexicon;
+  external Pointer<Utf8> ruleFsts;
+}
+
 final class SherpaOnnxOnlineRecognizerConfig extends Struct {
   external SherpaOnnxFeatureConfig feat;
   external SherpaOnnxOnlineModelConfig model;
@@ -441,6 +448,7 @@ final class SherpaOnnxOnlineRecognizerConfig extends Struct {
 
   @Int32()
   external int hotwordsBufSize;
+  external SherpaOnnxHomophoneReplacerConfig hr;
 }
 
 final class SherpaOnnxSileroVadModelConfig extends Struct {