vosk-flutter
vosk-flutter copied to clipboard
How to add multiple models to the app and use it at once so that it improves accuracy and helpful for wide range of users
import 'dart:async'; import 'dart:convert'; import 'dart:developer'; import 'dart:io'; import 'dart:typed_data'; import 'package:edolms/modules/screens/student/ai/conversation/chat/functions/model/vosk_result_model.dart'; import 'package:flutter/services.dart'; import 'package:vosk_flutter/vosk_flutter.dart';
class VoskTranscriber { late VoskFlutterPlugin _vosk; Recognizer? _recognizer; SpeechService? _speechService; Model? _model; String? _error; late StreamSubscription<String>? _resultSubscription;
VoskTranscriber() { initializeVosk(); }
Future
print("_initializeVosk called");
_vosk = VoskFlutterPlugin.instance();
final modelPath = await ModelLoader()
.loadFromAssets('assets/models/vosk-model-small-en-in-0.4.zip')
.then(
(modelPath) => _vosk.createModel(modelPath)) // create model object
.then((model) => _model = model)
.then((_) => _vosk.createRecognizer(
model: _model!, sampleRate: _sampleRate)) // create recognizer
.then((value) => _recognizer = value)
.then((recognizer) {
if (Platform.isAndroid) {
_vosk
.initSpeechService(_recognizer!) // init speech service
.then((speechService) => _speechService = speechService)
.catchError((e) => _error = e.toString());
print("sampleRate : ${_sampleRate}");
}
}).catchError((e) async {
_error = e.toString();
print("catchError : $e");
return null;
});
// print("modelPath :$modelPath");
// _recognizer = await _vosk.createRecognizer(
// model: Model(modelPath, const MethodChannel("")), sampleRate: 16000);
}
Future<String> transcribeAudio(Uint8List audioBytes) async { print("transcribeAudio function called"); print("audioBytes length : ${audioBytes.length}"); List<String> results = []; int chunkSize = 348135; int pos = 0;
while (pos + chunkSize < audioBytes.length) {
final resultReady = await _recognizer!.acceptWaveformBytes(
Uint8List.fromList(audioBytes.sublist(pos, pos + chunkSize)),
);
// print("resultReady : ${resultReady}");
pos += chunkSize;
if (resultReady) {
String result = await _recognizer!.getResult();
dynamic decodedData = jsonDecode(result);
if (decodedData["text"].isEmpty ||
decodedData["text"] == "" ||
decodedData["text"] == null) {
print(" transcribed text empty ");
} else {
log("complete result in vosk : ${result}");
results.add(result);
}
} else {
String result = await _recognizer!.getPartialResult();
dynamic decodedData = jsonDecode(result);
if (decodedData["partial"].isEmpty ||
decodedData["partial"] == "" ||
decodedData["partial"] == null) {
print("empty partial result");
} else {
log("partial result in vosk : ${result}");
results.add(result);
}
}
}
// await _recognizer!
// .acceptWaveformBytes(Uint8List.fromList(audioBytes.sublist(pos)));
results.add(await _recognizer!.getFinalResult());
return results.join(' ');
}
startSpeechToText(Function(String) onResult) async { String result = "";
// final speechService = await _vosk.initSpeechService(_recognizer!);
_speechService?.onPartial().forEach((partial) => print(partial));
_speechService?.onResult().forEach((result) => print(result));
await _speechService?.start();
_resultSubscription = _speechService?.onResult().listen((event) {
result = _onSpeechResult(event);
onResult(result);
});
}
String _onSpeechResult(dynamic result) { String _sttAllWords = ""; Map<String, dynamic> decoded = json.decode(result); VoskResultModel voskResult = VoskResultModel.fromJson(decoded); if (voskResult.text != null && voskResult.text != "") { _sttAllWords += voskResult.text?.trim() ?? ""; print("final text: ${_sttAllWords.trim()}"); }
return _sttAllWords;
}
Future
print("stopped speechToText wosk");
} }
@alejandrogiubel @sergsavchuk @nshmyrev @EvgeniyVabishchevich