fix(voice-call): Improve async handling and state management

Refactor voice call service to handle asynchronous operations more
precisely. Update method signatures to be async, use unawaited for
non-blocking calls, and ensure proper state reset between sessions.
Improve error handling and resource management for voice input and
text-to-speech services.

lib/features/chat/services/text_to_speech_service.dart

@@ -149,6 +149,9 @@ class TextToSpeechService {
   Future<void> dispose() async {
     await _eventSubscription?.cancel();
     _eventSubscription = null;
+
+    // Reset the singleton state for next session
+    await TtsManager.instance.reset();
   }
 
   /// Updates TTS settings.

lib/features/chat/services/tts_manager.dart

@@ -376,6 +376,28 @@ class TtsManager {
     }
   }
 
+  /// Resets the manager state for a new session.
+  ///
+  /// Call this between voice calls to ensure clean state. This clears
+  /// playback buffers and resets session tracking without destroying
+  /// the singleton instance.
+  Future<void> reset() async {
+    await stop();
+
+    // Reset playback state
+    _resetPlaybackState();
+    _activeSession = null;
+    _sessionCounter = 0;
+
+    // Reset server audio buffer
+    _serverAudioBuffer.clear();
+    _serverWaitingForNext = false;
+
+    // Reset cached voice defaults so they're refetched if needed
+    _serverDefaultVoice = null;
+    _serverDefaultVoiceFuture = null;
+  }
+
   /// Disposes the manager and releases resources.
   Future<void> dispose() async {
     await stop();

lib/features/chat/services/voice_call_service.dart

@@ -395,7 +395,7 @@ class VoiceCallService {
       _activeAssistantMessageId = null;
       _responseCompleted = false;
       _listeningSuspendedForSpeech = false;
-      _resetServerAudio(stopPlayback: true);
+      await _resetServerAudio(stopPlayback: true);
 
       if (_pauseReasons.isNotEmpty) {
         _listeningPaused = true;
@@ -584,7 +584,11 @@ class VoiceCallService {
     _speechQueue.clear();
     _enqueuedSentenceCount = 0;
     _responseCompleted = false;
-    _resetServerAudio(stopPlayback: true);
+    // Fire-and-forget: This is a synchronous event handler where blocking
+    // would delay socket event processing. The audio player stop is fast
+    // and race conditions are acceptable here since new playback will
+    // wait for this session to complete anyway.
+    unawaited(_resetServerAudio(stopPlayback: true));
     if (_isSpeaking) {
       _isSpeaking = false;
       unawaited(_tts.stop());
@@ -749,14 +753,14 @@ class VoiceCallService {
     _maybeResumeListeningAfterSpeech();
   }
 
-  void _resetServerAudio({bool stopPlayback = false}) {
+  Future<void> _resetServerAudio({bool stopPlayback = false}) async {
     _serverAudioBuffer.clear();
     _pendingServerAudioFetches = 0;
     _serverAudioSession++;
     _nextServerChunkId = 0;
     _nextServerPlaybackId = 0;
     if (stopPlayback) {
-      unawaited(_serverAudioPlayer.stop());
+      await _serverAudioPlayer.stop();
       _isSpeaking = false;
     }
     _serverPipelineActive = false;
@@ -819,7 +823,7 @@ class VoiceCallService {
     if (_isDisposed) return;
     _isSpeaking = false;
     _speechQueue.clear();
-    _resetServerAudio(stopPlayback: true);
+    unawaited(_resetServerAudio(stopPlayback: true));
     _listeningSuspendedForSpeech = false;
     _updateState(VoiceCallState.error);
     // Try to recover by restarting listening
@@ -865,7 +869,7 @@ class VoiceCallService {
     _listeningSuspendedForSpeech = false;
     _activeAssistantMessageId = null;
     _isSpeaking = false;
-    _resetServerAudio(stopPlayback: true);
+    await _resetServerAudio(stopPlayback: true);
     _updateState(VoiceCallState.disconnected);
   }
 
@@ -911,7 +915,7 @@ class VoiceCallService {
     _enqueuedSentenceCount = 0;
     _responseCompleted = false;
     _listeningSuspendedForSpeech = false;
-    _resetServerAudio(stopPlayback: true);
+    await _resetServerAudio(stopPlayback: true);
     await _tts.stop();
     _isSpeaking = false;
     _accumulatedResponse = '';
@@ -985,7 +989,7 @@ class VoiceCallService {
       _enqueuedSentenceCount = 0;
       _responseCompleted = false;
       _listeningSuspendedForSpeech = false;
-      _resetServerAudio(stopPlayback: true);
+      unawaited(_resetServerAudio(stopPlayback: true));
       pauseListening(reason: VoiceCallPauseReason.mute);
     } else {
       resumeListening(reason: VoiceCallPauseReason.mute);
@@ -1006,7 +1010,7 @@ class VoiceCallService {
     _callKitEventSubscription = null;
     _socketSubscription?.dispose();
 
-    _voiceInput.dispose();
+    await _voiceInput.dispose();
     await _tts.dispose();
     await _serverAudioStateSub?.cancel();
     await _serverAudioPlayer.dispose();
@@ -1029,7 +1033,7 @@ class VoiceCallService {
   }
 }
 
-@Riverpod(keepAlive: true)
+@riverpod
 VoiceCallService voiceCallService(Ref ref) {
   final voiceInput = ref.watch(voiceInputServiceProvider);
   final api = ref.watch(apiServiceProvider);

lib/features/chat/services/voice_input_service.dart

@@ -1026,12 +1026,12 @@ class VoiceInputService {
     });
   }
 
-  void dispose() {
-    stopListening();
-    unawaited(_disposeVadHandler());
-    unawaited(_microphonePermissionProbe.dispose());
+  Future<void> dispose() async {
+    await stopListening();
+    await _disposeVadHandler();
+    await _microphonePermissionProbe.dispose();
     try {
-      _speech.stop();
+      await _speech.stop();
     } catch (_) {}
   }
 }