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

[kuberanb]

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? _resultSubscription;

VoskTranscriber() { initializeVosk(); }

Future initializeVosk() async { const _sampleRate = 16000;

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 transcribeAudio(Uint8List audioBytes) async { print("transcribeAudio function called"); print("audioBytes length : ${audioBytes.length}"); List 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 stopSpeechToText() async { await _resultSubscription?.cancel(); await _speechService?.stop();

print("stopped speechToText wosk");

} }

[kuberanb]

@alejandrogiubel @sergsavchuk @nshmyrev @EvgeniyVabishchevich