deepvoice_client/lib/data/services/azure_asr_service.dart

import 'dart:async';
import 'package:flutter/services.dart';
import 'package:flutter_dotenv/flutter_dotenv.dart';
import '../../../core/utils/logger.dart';
import 'package:get/get.dart';
/// 微软语音识别服务
/// 
/// 该服务提供了通过平台通道与 Android 上的 Microsoft Speech SDK 交互的接口
class AzureAsrService {
  static final AzureAsrService to = Get.put(AzureAsrService());
  static const MethodChannel _channel = MethodChannel('com.deep_voice.azure_asr');
  static const EventChannel _eventChannel = EventChannel('com.deep_voice.azure_asr_events');
  
  bool _isInitialized = false;
  late final String _subscriptionKey;
  late final String _serviceRegion;
  

  List<String> _supportedLanguages = ['zh-CN', 'en-US'];
  List<String> get supportedLanguages => _supportedLanguages;
  
  // 连续识别相关
  bool _isContinuousRecognitionActive = false;
  StreamController<RecognitionEvent>? _eventStreamController;
  StreamSubscription? _eventSubscription;
  
  // 事件通道状态
  bool _isEventChannelReady = false;
  Completer<void>? _eventChannelReadyCompleter;
  
  // 公开的事件流
  Stream<RecognitionEvent>? _recognitionStream;
  Stream<RecognitionEvent>? get recognitionStream => _recognitionStream;
  
  // 最新的识别结果
  String _latestRecognizedText = '';
  String get latestRecognizedText => _latestRecognizedText;
  
  // 最新检测到的语言
  String _latestDetectedLanguage = '';
  String get latestDetectedLanguage => _latestDetectedLanguage;
  
  AzureAsrService() {
    _loadConfig();
    _setupEventChannel();
  }
  
  /// 从环境变量加载配置
  void _loadConfig() {
    _subscriptionKey = dotenv.env['AZURE_SPEECH_KEY'] ?? '';
    _serviceRegion = dotenv.env['AZURE_SPEECH_REGION'] ?? '';
    
    if (_subscriptionKey.isEmpty || _serviceRegion.isEmpty) {
      throw Exception('未找到 Azure 语音服务配置。请在 .env 文件中设置 AZURE_SPEECH_KEY 和 AZURE_SPEECH_REGION');
    }
  }
  
  /// 设置事件通道
  void _setupEventChannel() {
    _eventChannelReadyCompleter = Completer<void>();
    
    _eventSubscription = _eventChannel
        .receiveBroadcastStream()
        .listen((event) {
          if (event is Map) {
            final String eventType = event['type'] as String? ?? '';
            
            // 处理通道准备好的事件
            if (eventType == 'channelReady') {
              _isEventChannelReady = true;
              if (!_eventChannelReadyCompleter!.isCompleted) {
                _eventChannelReadyCompleter!.complete();
              }
              return;
            }
            
            _handleRecognitionEvent(event);
          }
        }, onError: _handleRecognitionError);
  }
  
  /// 等待事件通道准备好
  Future<bool> _waitForEventChannel({Duration timeout = const Duration(seconds: 5)}) async {
    if (_isEventChannelReady) return true;
    
    try {
      await _eventChannelReadyCompleter!.future.timeout(timeout);
      return true;
    } on TimeoutException {
      Logger.error('等待事件通道准备好超时');
      return false;
    }
  }
  
  /// 初始化微软语音识别 SDK
  /// 
  /// 返回 true 表示初始化成功，否则抛出 PlatformException
  Future<bool> initialize({
    List<String>? supportedLanguages,
  }) async {
    try {

      if (supportedLanguages != null && supportedLanguages.isNotEmpty) {
        _supportedLanguages = supportedLanguages;
      }
      
      // 检查是否需要重新初始化
      if (_isInitialized) {
        await dispose();
      }
      
      final bool result = await _channel.invokeMethod('initialize', {
        'subscriptionKey': _subscriptionKey,
        'region': _serviceRegion,
        'supportedLanguages': _supportedLanguages,
      });
      
      _isInitialized = result;
      Logger.info('Azure 语音识别服务初始化${result ? '成功' : '失败'}');
      return result;
    } catch (e) {
      Logger.error('Azure 语音识别服务初始化失败: $e');
      _isInitialized = false;
      rethrow;
    }
  }
  
  
  /// 执行一次性语音识别
  /// 
  /// 返回识别的文本和检测到的语言，如果识别失败则抛出 PlatformException
  Future<RecognitionResult> recognizeSpeech() async {
    if (!_isInitialized) {
      await initialize();
    }
    
    try {
      final result = await _channel.invokeMethod('recognizeOnce');
      
      if (result is Map) {
        final String text = result['text'] as String? ?? '';
        final String detectedLanguage = result['detectedLanguage'] as String? ?? '';
        _latestRecognizedText = text;
        _latestDetectedLanguage = detectedLanguage;
        return RecognitionResult(text: text, detectedLanguage: detectedLanguage);
      } else if (result is String) {
        // 兼容旧版本
        _latestRecognizedText = result;
        return RecognitionResult(text: result);
      }
      
      throw Exception('无效的识别结果格式');
    } catch (e) {
      Logger.error('语音识别失败: $e');
      rethrow;
    }
  }
  
  /// 开始连续语音识别
  /// 
  /// 返回一个包含识别事件的流，如果开始失败则抛出 PlatformException
  Future<Stream<RecognitionEvent>> startContinuousRecognition() async {
    if (!_isInitialized) {
      await initialize();
    }
    
    // 确保事件通道已准备好
    if (!await _waitForEventChannel()) {
      throw Exception('事件通道未准备好，无法开始连续识别');
    }
    
    if (_isContinuousRecognitionActive) {
      await stopContinuousRecognition();
    }
    
    try {
      _eventStreamController = StreamController<RecognitionEvent>.broadcast();
      
      // 开始连续识别
      final bool result = await _channel.invokeMethod('startContinuousRecognition');
      
      if (!result) {
        _cleanupEventStream();
        throw Exception('启动连续识别失败');
      }
      
      _isContinuousRecognitionActive = true;
      return _eventStreamController!.stream;
    } catch (e) {
      Logger.error('开始连续语音识别失败: $e');
      _cleanupEventStream();
      rethrow;
    }
  }
  
  /// 停止连续语音识别
  /// 
  /// 返回 true 表示停止成功，否则抛出 PlatformException
  Future<bool> stopContinuousRecognition() async {
    if (!_isInitialized || !_isContinuousRecognitionActive) {
      return true;
    }
    
    try {
      final bool result = await _channel.invokeMethod('stopContinuousRecognition');
      _cleanupEventStream();
      return result;
    } catch (e) {
      Logger.error('停止连续语音识别失败: $e');
      _cleanupEventStream();
      rethrow;
    }
  }
  
  /// 检查连续识别是否活跃
  bool isContinuousRecognitionActive() {
    return _isContinuousRecognitionActive;
  }
  
  /// 处理来自原生端的识别事件
  void _handleRecognitionEvent(dynamic event) {
    if (event is! Map || _eventStreamController == null) return;
    
    final Map<dynamic, dynamic> eventMap = event;
    final String eventType = eventMap['type'] as String? ?? '';
    
    switch (eventType) {
      case 'result':
        final String text = eventMap['text'] as String? ?? '';
        final String detectedLanguage = eventMap['detectedLanguage'] as String? ?? '';
        _latestRecognizedText = text;
        _latestDetectedLanguage = detectedLanguage;
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.finalResult,
          text: text,
          detectedLanguage: detectedLanguage,
        ));
        break;
        
      case 'recognizing':
        final String text = eventMap['text'] as String? ?? '';
        final String detectedLanguage = eventMap['detectedLanguage'] as String? ?? '';
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.intermediateResult,
          text: text,
          detectedLanguage: detectedLanguage,
        ));
        break;
        
      case 'sessionStarted':
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.sessionStarted,
        ));
        break;
        
      case 'sessionStopped':
        _isContinuousRecognitionActive = false;
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.sessionStopped,
        ));
        break;
        
      case 'canceled':
        _isContinuousRecognitionActive = false;
        final String reason = eventMap['reason'] as String? ?? '';
        final String errorDetails = eventMap['errorDetails'] as String? ?? '';
        
        if (reason.isNotEmpty || errorDetails.isNotEmpty) {
          Logger.error('识别取消: $reason - $errorDetails');
        }
        
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.canceled,
          error: '$reason: $errorDetails',
        ));
        break;
        
      case 'error':
        final String error = eventMap['message'] as String? ?? '';
        Logger.error('识别错误: $error');
        _eventStreamController?.add(RecognitionEvent(
          type: RecognitionEventType.error,
          error: error,
        ));
        break;
    }
  }
  
  /// 处理识别事件流错误
  void _handleRecognitionError(Object error) {
    Logger.error('识别事件流错误: $error');
    _eventStreamController?.addError(error);
    _cleanupEventStream();
  }
  
  /// 清理事件流资源
  void _cleanupEventStream() {
    _eventStreamController?.close();
    _eventStreamController = null;
    _isContinuousRecognitionActive = false;
  }
  
  /// 释放资源
  Future<void> dispose() async {
    try {
      if (_isContinuousRecognitionActive) {
        await stopContinuousRecognition();
      }
      
      // 通知原生端释放资源
      await _channel.invokeMethod('dispose');
      _isInitialized = false;
      
      Logger.info('Azure 语音识别资源已释放');
    } catch (e) {
      Logger.error('释放语音识别资源失败: $e');
      _cleanupEventStream();
      _isInitialized = false;
    }
  }

}

/// 识别事件类型
enum RecognitionEventType {
  /// 最终识别结果
  finalResult,
  
  /// 中间识别结果（实时反馈）
  intermediateResult,
  
  /// 会话开始
  sessionStarted,
  
  /// 会话结束
  sessionStopped,
  
  /// 识别取消
  canceled,
  
  /// 识别错误
  error,
}

/// 识别事件
class RecognitionEvent {
  /// 事件类型
  final RecognitionEventType type;
  
  /// 识别文本（仅在 finalResult 和 intermediateResult 类型中有效）
  final String text;
  
  /// 检测到的语言（仅在启用自动语言检测时有效）
  final String detectedLanguage;
  
  /// 错误信息（仅在 error 和 canceled 类型中有效）
  final String error;
  
  RecognitionEvent({
    required this.type,
    this.text = '',
    this.detectedLanguage = '',
    this.error = '',
  });
  
  @override
  String toString() {
    return 'RecognitionEvent{type: $type, text: $text, detectedLanguage: $detectedLanguage, error: $error}';
  }
}

/// 识别结果
class RecognitionResult {
  /// 识别文本
  final String text;
  
  /// 检测到的语言（仅在启用自动语言检测时有效）
  final String detectedLanguage;
  
  RecognitionResult({
    required this.text,
    this.detectedLanguage = '',
  });
  
  @override
  String toString() {
    return 'RecognitionResult{text: $text, detectedLanguage: $detectedLanguage}';
  }
}