Flutter/Dart：任何语言的语音到文本（离线和连续）

Question

我可以使用任何软件包来创建可以将语音转文本的应用程序吗？

它应该包括以下功能：

• 离线语音转文字
• 连续聆听（10-30 分钟）
• 识别器适用于 Android/IOS 支持的所有语言

到目前为止，我发现了这个https://pub.dev/packages/speech_recognition，但它说：

iOS API 发送中间结果，在我的 Android 设备上，只有 最终转录已收到

其他限制：在 iOS 上，默认情况下插件配置为 法语、英语、俄语、西班牙语、意大利语。在 Android 上，没有 额外的安装，它可能只适用于默认 设备语言环境。

有人测试过这个包，结果很好？或者您有什么其他建议？

Answer 1

我现在使用https://pub.dev/packages/speech_to_text。它得到了积极的维护，而且效果很好。我认为可以编写一些自定义代码来让它连续监听。

编辑：

根据要求，请参阅下面的连续收听逻辑。我仅将其用作概念证明，因此我不建议将其用于生产应用程序。据我所知，Android API 不支持开箱即用的连续监听。

语音识别块

import 'package:bloc/bloc.dart';
import 'package:meta/meta.dart';
import 'package:template_mobile/core/sevices/speech_recognition_service.dart';
import 'package:template_mobile/core/state/event/speech_recognition_event.dart';
import 'package:template_mobile/core/state/state/speech_recognition_state.dart';

class SpeechRecognitionBloc
    extends Bloc<SpeechRecognitionEvent, SpeechRecognitionState> {
  final SpeechRecognitionService speechRecognitionService;

  SpeechRecognitionBloc({
    @required this.speechRecognitionService,
  }) : assert(speechRecognitionService != null) {
    speechRecognitionService.errors.stream.listen((errorResult) {
      add(SpeechRecognitionErrorEvent(
        error: "${errorResult.errorMsg} - ${errorResult.permanent}",
      ));
    });

    speechRecognitionService.statuses.stream.listen((status) {
      if (state is SpeechRecognitionRecognizedState) {
        var currentState = state as SpeechRecognitionRecognizedState;
        if (currentState.finalResult) {
          add(SpeechRecognitionStatusChangedEvent());
        }
      }
    });

    speechRecognitionService.words.stream.listen((speechResult) {
      add(SpeechRecognitionRecognizedEvent(
        words: speechResult.recognizedWords,
        finalResult: speechResult.finalResult,
      ));
    });
  }

  @override
  SpeechRecognitionState get initialState =>
      SpeechRecognitionUninitializedState();

  @override
  Stream<SpeechRecognitionState> mapEventToState(
      SpeechRecognitionEvent event) async* {
    if (event is SpeechRecognitionInitEvent) {
      var hasSpeech = await speechRecognitionService.initSpeech();
      if (hasSpeech) {
        yield SpeechRecognitionAvailableState();
      } else {
        yield SpeechRecognitionUnavailableState();
      }
    }

    if (event is SpeechRecognitionStartPressEvent) {
      yield SpeechRecognitionStartPressedState();
      add(SpeechRecognitionStartEvent());
    }

    if (event is SpeechRecognitionStartEvent) {
      speechRecognitionService.startListening();
      yield SpeechRecognitionStartedState();
    }

    if (event is SpeechRecognitionStopPressEvent) {
      yield SpeechRecognitionStopPressedState();
      add(SpeechRecognitionStopEvent());
    }

    if (event is SpeechRecognitionStopEvent) {
      speechRecognitionService.stopListening();
      yield SpeechRecognitionStopedState();
    }

    if (event is SpeechRecognitionCancelEvent) {
      speechRecognitionService.cancelListening();
      yield SpeechRecognitionCanceledState();
    }

    if (event is SpeechRecognitionRecognizedEvent) {
      yield SpeechRecognitionRecognizedState(
          words: event.words, finalResult: event.finalResult);
      if (event.finalResult == true &&
          speechRecognitionService.statuses.value == 'notListening') {
        await Future.delayed(Duration(milliseconds: 50));
        add(SpeechRecognitionStatusChangedEvent());
      }
    }

    if (event is SpeechRecognitionErrorEvent) {
      yield SpeechRecognitionErrorState(error: event.error);
      // Just for UI updates for the state to propagates
      await Future.delayed(Duration(milliseconds: 50));
      add(SpeechRecognitionInitEvent());
      await Future.delayed(Duration(milliseconds: 50));
      add(SpeechRecognitionStartPressEvent());
    }

    if (event is SpeechRecognitionStatusChangedEvent) {
      yield SpeechRecognitionStatusState();
      add(SpeechRecognitionStartPressEvent());
    }
  }
}

语音识别服务

import 'dart:async';

import 'package:rxdart/rxdart.dart';
import 'package:speech_to_text/speech_recognition_error.dart';
import 'package:speech_to_text/speech_recognition_result.dart';
import 'package:speech_to_text/speech_to_text.dart';

class SpeechRecognitionService {
  final SpeechToText speech = SpeechToText();

  var errors = StreamController<SpeechRecognitionError>();
  var statuses = BehaviorSubject<String>();
  var words = StreamController<SpeechRecognitionResult>();

  var _localeId = '';

  Future<bool> initSpeech() async {
    bool hasSpeech = await speech.initialize(
      onError: errorListener,
      onStatus: statusListener,
    );

    if (hasSpeech) {
      var systemLocale = await speech.systemLocale();
      _localeId = systemLocale.localeId;
    }

    return hasSpeech;
  }

  void startListening() {
    speech.stop();
    speech.listen(
        onResult: resultListener,
        listenFor: Duration(minutes: 1),
        localeId: _localeId,
        onSoundLevelChange: null,
        cancelOnError: true,
        partialResults: true);
  }

  void errorListener(SpeechRecognitionError error) {
    errors.add(error);
  }

  void statusListener(String status) {
    statuses.add(status);
  }

  void resultListener(SpeechRecognitionResult result) {
    words.add(result);
  }

  void stopListening() {
    speech.stop();
  }

  void cancelListening() {
    speech.cancel();
  }
}

Answer 2

https://pub.dev/packages/speech_recognition 是最好的选择。 它基于SpeechRecognizer 并提供离线语音到文本。

无法连续收听。即使是付费的在线Cloud Speech-to-Text API 也不允许这样做，因为它很危险（滥用等）。

在 iOS 上，默认情况下插件配置为法语、英语、俄语、西班牙语、意大利语，但您可以将缺少的语言添加到 swift 源文件中。

所以最终你不会找到更好的语音识别插件，即使它并不完美。

Answer 3

如果您愿意编写自定义平台特定代码https://flutter.dev/docs/development/platform-integration/platform-channels，对于Android，您可以使用https://github.com/alphacep/vosk-android-demo 来获得您想要的。

另一种方法是按照How Can I Implement Google Voice Typing In My Application? 中的建议构建您自己的键盘输入法以使用语音输入。