feat: mvp viode
This commit is contained in:
liqupan
303
wei_ai_app/lib/screens/interaction/voice_session_controller.dart
Normal file
303
wei_ai_app/lib/screens/interaction/voice_session_controller.dart
Normal file
@@ -0,0 +1,303 @@
|
||||
import 'dart:async';
|
||||
import 'package:flutter/foundation.dart';
|
||||
import '../../core/core.dart';
|
||||
import 'package:permission_handler/permission_handler.dart';
|
||||
|
||||
enum VoiceState {
|
||||
listening, // Waiting for user input
|
||||
processing, // Sending to AI
|
||||
speaking, // AI is talking
|
||||
error,
|
||||
}
|
||||
|
||||
class VoiceSessionController extends ChangeNotifier {
|
||||
final CharacterModel character;
|
||||
final Function(String) onUserMessage;
|
||||
final Function(ChatMessage) onAiMessage;
|
||||
|
||||
// Speaking Queue
|
||||
final List<String> _speakQueue = [];
|
||||
bool _isSpeaking = false;
|
||||
Timer? _silenceTimer;
|
||||
|
||||
VoiceState _state = VoiceState.listening;
|
||||
String _recognizedText = '';
|
||||
String _aiTypingText = '';
|
||||
bool _isMicMuted = false;
|
||||
|
||||
// Services
|
||||
final STTService _stt = STTService();
|
||||
final TTSService _tts = TTSService();
|
||||
|
||||
// State getters
|
||||
VoiceState get state => _state;
|
||||
String get recognizedText => _recognizedText;
|
||||
String get aiTypingText => _aiTypingText;
|
||||
bool get isMicMuted => _isMicMuted;
|
||||
|
||||
// Buffer for sentence completion
|
||||
String _sentenceBuffer = '';
|
||||
final List<String> _punctuation = ['。', '？', '！', '.', '?', '!', '\n'];
|
||||
|
||||
VoiceSessionController({
|
||||
required this.character,
|
||||
required this.onUserMessage,
|
||||
required this.onAiMessage,
|
||||
}) {
|
||||
_init();
|
||||
}
|
||||
|
||||
Future<void> _init() async {
|
||||
// Request permissions
|
||||
await [Permission.microphone, Permission.speech].request();
|
||||
|
||||
// Init services
|
||||
await _stt.init();
|
||||
await _tts.init();
|
||||
|
||||
// Setup TTS callbacks
|
||||
_tts.setStartHandler(() {
|
||||
debugPrint('🔊 TTS Started');
|
||||
// Already paused STT in _processSpeakQueue
|
||||
});
|
||||
|
||||
_tts.setCompletionHandler(() {
|
||||
debugPrint('✅ TTS Completed');
|
||||
_isSpeaking = false;
|
||||
_processSpeakQueue(); // Play next
|
||||
});
|
||||
|
||||
_tts.setErrorHandler((msg) {
|
||||
debugPrint('❌ TTS Error: $msg');
|
||||
_isSpeaking = false;
|
||||
_processSpeakQueue();
|
||||
});
|
||||
|
||||
// Start listening immediately
|
||||
if (!_isMicMuted) {
|
||||
startListening();
|
||||
}
|
||||
}
|
||||
|
||||
void toggleMic() {
|
||||
_isMicMuted = !_isMicMuted;
|
||||
if (_isMicMuted) {
|
||||
stopListening();
|
||||
} else {
|
||||
if (_state == VoiceState.listening) {
|
||||
startListening();
|
||||
}
|
||||
}
|
||||
notifyListeners();
|
||||
}
|
||||
|
||||
Future<void> startListening() async {
|
||||
if (_isMicMuted) return;
|
||||
|
||||
_state = VoiceState.listening;
|
||||
_recognizedText = '';
|
||||
_lastProcessedLength = 0;
|
||||
notifyListeners();
|
||||
|
||||
// Stop TTS if it's playing (Interruption)
|
||||
if (_isSpeaking || _speakQueue.isNotEmpty) {
|
||||
_speakQueue.clear();
|
||||
await _tts.stop();
|
||||
_isSpeaking = false;
|
||||
}
|
||||
|
||||
await _stt.listen(
|
||||
onResult: (text) {
|
||||
_recognizedText = text;
|
||||
notifyListeners();
|
||||
},
|
||||
onFinalResult: (text) {
|
||||
_recognizedText = text;
|
||||
notifyListeners();
|
||||
_processUserMessage(text);
|
||||
},
|
||||
localeId: 'zh-CN', // Make dynamic later if needed
|
||||
);
|
||||
}
|
||||
|
||||
Future<void> stopListening() async {
|
||||
await _stt.stop();
|
||||
}
|
||||
|
||||
Future<void> _processUserMessage(String text) async {
|
||||
if (text.trim().isEmpty) {
|
||||
// If empty, just listen again
|
||||
startListening();
|
||||
return;
|
||||
}
|
||||
|
||||
_state = VoiceState.processing;
|
||||
onUserMessage(text); // Notify UI to show user message
|
||||
notifyListeners();
|
||||
|
||||
// Construct history for context (simplified for now, ideally pass full history)
|
||||
// We will rely on ChatService to handle the full history if we pass the latest message
|
||||
// But ChatService needs the list. For the voice mode, let's assume interacting adds to DB
|
||||
// and we might need to fetch fresh context or pass it in.
|
||||
// Ideally, the InteractionScreen manages the source of truth for messages.
|
||||
// Here we'll just send the text to prompt the AI.
|
||||
|
||||
// Construct history
|
||||
final session = await ChatStorageService.getSession(character.id);
|
||||
var messages = session.messages;
|
||||
|
||||
// ChatService.sendMessage appends the userMessage automatically.
|
||||
// We need to ensure 'messages' doesn't already contain it (if DB write was fast).
|
||||
if (messages.isNotEmpty) {
|
||||
final lastMsg = messages.last;
|
||||
if (lastMsg.isUser && lastMsg.content == text) {
|
||||
messages = List.from(messages)..removeLast();
|
||||
}
|
||||
}
|
||||
|
||||
_aiTypingText = '';
|
||||
_sentenceBuffer = '';
|
||||
_lastProcessedLength = 0;
|
||||
|
||||
try {
|
||||
final fullResponse = await ChatService.sendMessage(
|
||||
character: character,
|
||||
messages: messages,
|
||||
userMessage: text, // ChatService handles appending this if we use the right method
|
||||
onStream: (content) {
|
||||
_aiTypingText = content;
|
||||
_processStreamChunk(content);
|
||||
notifyListeners();
|
||||
},
|
||||
);
|
||||
|
||||
// Process any remaining text in buffer
|
||||
if (_sentenceBuffer.isNotEmpty) {
|
||||
if (_state != VoiceState.speaking) {
|
||||
_state = VoiceState.speaking;
|
||||
notifyListeners();
|
||||
}
|
||||
await _speak(_sentenceBuffer);
|
||||
}
|
||||
|
||||
// Interaction finished, save AI message
|
||||
final aiMsg = ChatMessage.assistant(fullResponse);
|
||||
onAiMessage(aiMsg);
|
||||
|
||||
// Note: We do NOT immediately startListening here.
|
||||
// We rely on the TTS Completion Handler to trigger startListening
|
||||
// when the entire queue is drained.
|
||||
|
||||
} catch (e) {
|
||||
debugPrint('❌ Voice Process Error: $e');
|
||||
_state = VoiceState.error;
|
||||
notifyListeners();
|
||||
// Retry listening after error
|
||||
Future.delayed(const Duration(seconds: 2), startListening);
|
||||
}
|
||||
}
|
||||
|
||||
// Better implementation needs to handle state to avoid infinite loops
|
||||
int _lastProcessedLength = 0;
|
||||
|
||||
void _processStreamChunk(String content) {
|
||||
if (_state != VoiceState.speaking) {
|
||||
_state = VoiceState.speaking;
|
||||
notifyListeners();
|
||||
}
|
||||
|
||||
// Calculate delta (new content only)
|
||||
if (content.length <= _lastProcessedLength) return;
|
||||
|
||||
String delta = content.substring(_lastProcessedLength);
|
||||
_lastProcessedLength = content.length;
|
||||
_sentenceBuffer += delta;
|
||||
|
||||
// Check for punctuation to split sentences
|
||||
bool foundPunctuation = false;
|
||||
for (var p in _punctuation) {
|
||||
if (_sentenceBuffer.contains(p)) {
|
||||
foundPunctuation = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (foundPunctuation) {
|
||||
_processBufferForSentences();
|
||||
}
|
||||
}
|
||||
|
||||
void _processBufferForSentences() {
|
||||
String tempBuffer = _sentenceBuffer;
|
||||
String keepBuffer = '';
|
||||
|
||||
// Simple tokenizer: split by punctuation but keep the punctuation attached to the sentence
|
||||
// This is a naive implementation.
|
||||
// "Hello! How are you?" -> ["Hello!", "How are you?"]
|
||||
|
||||
// We iterate through chars to find split points
|
||||
int lastSplitIndex = 0;
|
||||
|
||||
for (int i = 0; i < tempBuffer.length; i++) {
|
||||
String char = tempBuffer[i];
|
||||
if (_punctuation.contains(char)) {
|
||||
// Found end of a sentence
|
||||
String sentence = tempBuffer.substring(lastSplitIndex, i + 1);
|
||||
if (sentence.trim().isNotEmpty) {
|
||||
_speak(sentence);
|
||||
}
|
||||
lastSplitIndex = i + 1;
|
||||
}
|
||||
}
|
||||
|
||||
// Keep the remaining part that didn't end with punctuation
|
||||
if (lastSplitIndex < tempBuffer.length) {
|
||||
keepBuffer = tempBuffer.substring(lastSplitIndex);
|
||||
}
|
||||
|
||||
_sentenceBuffer = keepBuffer;
|
||||
}
|
||||
|
||||
Future<void> _speak(String text) async {
|
||||
// If we are listening (interrupted), ignore (or should check state)
|
||||
if (_state == VoiceState.listening) return;
|
||||
|
||||
_speakQueue.add(text);
|
||||
_processSpeakQueue();
|
||||
}
|
||||
|
||||
void _processSpeakQueue() async {
|
||||
if (_isSpeaking) return;
|
||||
|
||||
if (_speakQueue.isEmpty) {
|
||||
// All done speaking (or no TTS was produced)
|
||||
if (_state == VoiceState.speaking || _state == VoiceState.processing) {
|
||||
debugPrint('🎤 Queue empty, resuming listening...');
|
||||
_state = VoiceState.listening;
|
||||
notifyListeners();
|
||||
// Debounce STT restart to avoid rapid stop/start deadlocks
|
||||
_silenceTimer?.cancel();
|
||||
_silenceTimer = Timer(const Duration(milliseconds: 250), () {
|
||||
startListening();
|
||||
});
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
// Pop first
|
||||
String text = _speakQueue.removeAt(0);
|
||||
_isSpeaking = true;
|
||||
|
||||
// Ensure STT is paused while speaking
|
||||
await stopListening();
|
||||
|
||||
await _tts.speak(text);
|
||||
}
|
||||
|
||||
@override
|
||||
void dispose() {
|
||||
_stt.stop();
|
||||
_tts.stop();
|
||||
super.dispose();
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user