AI-Health/health_app/lib/services/realtime_speech_service.dart

import 'dart:async';
import 'dart:convert';
import 'dart:io';
import 'dart:typed_data';

import 'package:flutter_riverpod/flutter_riverpod.dart';
import 'package:record/record.dart';

import '../core/api_client.dart';
import '../providers/auth_provider.dart';

enum SpeechRecognitionEventType { partial, done, error }

class SpeechRecognitionEvent {
  final SpeechRecognitionEventType type;
  final String text;

  const SpeechRecognitionEvent(this.type, this.text);
}

class SpeechRecognitionException implements Exception {
  final String message;

  const SpeechRecognitionException(this.message);

  @override
  String toString() => message;
}

final realtimeSpeechServiceProvider = Provider<RealtimeSpeechService>((ref) {
  return RealtimeSpeechService(ref.watch(apiClientProvider));
});

class RealtimeSpeechService {
  final ApiClient _apiClient;

  const RealtimeSpeechService(this._apiClient);

  Future<bool> requestPermission() async {
    final recorder = AudioRecorder();
    try {
      return await recorder.hasPermission();
    } finally {
      recorder.dispose();
    }
  }

  Future<RealtimeSpeechSession> start() async {
    final token = await _apiClient.accessToken;
    if (token == null || token.isEmpty) {
      throw const SpeechRecognitionException('登录状态已失效，请重新登录');
    }

    final session = RealtimeSpeechSession(
      endpoint: buildRealtimeSpeechEndpoint(baseUrl),
      accessToken: token,
    );
    try {
      await session.start();
      return session;
    } catch (_) {
      await session.dispose();
      rethrow;
    }
  }
}

Uri buildRealtimeSpeechEndpoint(String apiBaseUrl) {
  final apiUri = Uri.parse(apiBaseUrl);
  final basePath = apiUri.path.endsWith('/')
      ? apiUri.path.substring(0, apiUri.path.length - 1)
      : apiUri.path;
  return apiUri.replace(
    scheme: apiUri.scheme == 'https' ? 'wss' : 'ws',
    path: '$basePath/api/speech/realtime',
    query: null,
    fragment: null,
  );
}

class RealtimeSpeechSession {
  final Uri endpoint;
  final String accessToken;
  final AudioRecorder _recorder = AudioRecorder();
  final StreamController<SpeechRecognitionEvent> _events =
      StreamController<SpeechRecognitionEvent>.broadcast();
  final Completer<void> _ready = Completer<void>();
  final Completer<String> _done = Completer<String>();

  WebSocket? _socket;
  StreamSubscription<Uint8List>? _audioSubscription;
  StreamSubscription<dynamic>? _socketSubscription;
  bool _finishing = false;
  bool _disposed = false;
  String _latestText = '';

  RealtimeSpeechSession({required this.endpoint, required this.accessToken}) {
    _ready.future.ignore();
    _done.future.ignore();
  }

  Stream<SpeechRecognitionEvent> get events => _events.stream;
  String get latestText => _latestText;

  Future<void> start() async {
    if (!await _recorder.hasPermission()) {
      throw const SpeechRecognitionException('没有麦克风权限，请在系统设置中允许后重试');
    }

    try {
      _socket = await WebSocket.connect(
        endpoint.toString(),
        headers: {'Authorization': 'Bearer $accessToken'},
      ).timeout(const Duration(seconds: 10));
    } on TimeoutException {
      throw const SpeechRecognitionException('连接语音服务超时，请检查网络');
    } on WebSocketException {
      throw const SpeechRecognitionException('无法连接语音服务，请稍后重试');
    }

    _socketSubscription = _socket!.listen(
      _handleSocketMessage,
      onError: _handleSocketError,
      onDone: _handleSocketDone,
      cancelOnError: false,
    );

    try {
      await _ready.future.timeout(const Duration(seconds: 10));
    } on TimeoutException {
      throw const SpeechRecognitionException('语音服务准备超时，请重试');
    }

    final audioStream = await _recorder.startStream(
      const RecordConfig(
        encoder: AudioEncoder.pcm16bits,
        sampleRate: 16000,
        numChannels: 1,
        autoGain: true,
        noiseSuppress: true,
        echoCancel: true,
        streamBufferSize: 3200,
      ),
    );
    _audioSubscription = audioStream.listen(
      (chunk) {
        final socket = _socket;
        if (!_finishing && socket?.readyState == WebSocket.open) {
          socket!.add(chunk);
        }
      },
      onError: (Object error, StackTrace stackTrace) {
        _completeError('麦克风录音失败，请重试');
      },
    );
  }

  Future<String> finish() async {
    if (_disposed) return _latestText;
    if (_finishing) return _done.future;
    _finishing = true;

    await _recorder.stop();
    await _audioSubscription?.cancel();
    _audioSubscription = null;

    // 后端可能在用户松手前已主动结束（例如达到 60 秒上限），
    // 此时 _done 已完成，直接返回结果即可，不需要再发 finish。
    if (_done.isCompleted) return _done.future;

    final socket = _socket;
    if (socket?.readyState != WebSocket.open) {
      throw const SpeechRecognitionException('语音连接已断开，请重试');
    }
    socket!.add(jsonEncode({'action': 'finish'}));

    try {
      return await _done.future.timeout(const Duration(seconds: 12));
    } on TimeoutException {
      throw const SpeechRecognitionException('识别结果返回超时，请重试');
    }
  }

  Future<void> cancel() async {
    if (_disposed) return;
    if (!_finishing) {
      _finishing = true;
      final socket = _socket;
      if (socket?.readyState == WebSocket.open) {
        socket!.add(jsonEncode({'action': 'cancel'}));
      }
    }
    await _recorder.cancel();
    await dispose();
  }

  void _handleSocketMessage(dynamic raw) {
    if (raw is! String) return;
    try {
      final decoded = jsonDecode(raw);
      if (decoded is! Map) return;
      final type = decoded['type']?.toString();
      if (type == 'ready') {
        if (!_ready.isCompleted) _ready.complete();
        return;
      }
      if (type == 'partial') {
        _latestText = decoded['text']?.toString() ?? _latestText;
        _events.add(
          SpeechRecognitionEvent(
            SpeechRecognitionEventType.partial,
            _latestText,
          ),
        );
        return;
      }
      if (type == 'done') {
        _latestText = decoded['text']?.toString() ?? _latestText;
        _events.add(
          SpeechRecognitionEvent(SpeechRecognitionEventType.done, _latestText),
        );
        if (!_done.isCompleted) _done.complete(_latestText);
        return;
      }
      if (type == 'error') {
        _completeError(decoded['message']?.toString() ?? '语音识别失败，请重试');
      }
    } catch (_) {
      _completeError('语音服务返回了无法识别的数据');
    }
  }

  void _handleSocketError(Object error, StackTrace stackTrace) {
    _completeError('语音连接异常，请检查网络');
  }

  void _handleSocketDone() {
    if (!_ready.isCompleted) {
      _ready.completeError(const SpeechRecognitionException('语音服务连接失败，请重试'));
    } else if (!_done.isCompleted && !_disposed) {
      _completeError('语音连接已断开，请重试');
    }
  }

  void _completeError(String message) {
    final error = SpeechRecognitionException(message);
    if (!_ready.isCompleted) _ready.completeError(error);
    if (!_done.isCompleted) _done.completeError(error);
    if (!_events.isClosed) {
      _events.add(
        SpeechRecognitionEvent(SpeechRecognitionEventType.error, message),
      );
    }
  }

  Future<void> dispose() async {
    if (_disposed) return;
    _disposed = true;
    await _audioSubscription?.cancel();
    await _socketSubscription?.cancel();
    final socket = _socket;
    if (socket != null && socket.readyState == WebSocket.open) {
      await socket.close(WebSocketStatus.normalClosure, 'completed');
    }
    await _events.close();
    _recorder.dispose();
  }
}