Update websocket_handler.py

websocket_handler.py

@@ -291,6 +291,10 @@ class RealtimeSession:
             if not self.is_streaming and self.is_websocket_active and self.state == ConversationState.LISTENING:
                 log_info(f"🔄 Restarting STT stream...", session_id=self.session.session_id)
                 
+                # Önce audio buffer'ı temizle - eski chunk'ları sil
+                await self.audio_buffer.clear()
+                self.chunk_counter = 0
+                
                 # Önce mevcut stream'i temizle (eğer varsa)
                 if self.stt_manager:
                     await self.stop_stt_streaming()
@@ -557,10 +561,14 @@ async def handle_audio_chunk(websocket: WebSocket, session: RealtimeSession, mes
         if session.state in [ConversationState.PLAYING_AUDIO, ConversationState.PROCESSING_TTS, 
                             ConversationState.PROCESSING_LLM, ConversationState.PROCESSING_STT]:
             log_debug(f"🔇 Ignoring audio chunk during state: {session.state.value}", session_id=session.session.session_id)
+            # Audio buffer'ı da temizle ki eski chunk'lar birikmesin
+            await session.audio_buffer.clear()
             return
         
         # Change state to listening if idle
         if session.state == ConversationState.IDLE:
+            # IDLE'dan LISTENING'e geçerken buffer'ı temizle
+            await session.audio_buffer.clear()
             await session.change_state(ConversationState.LISTENING)
             await websocket.send_json({
                 "type": "state_change",
@@ -604,6 +612,16 @@ async def handle_audio_chunk(websocket: WebSocket, session: RealtimeSession, mes
                 
                 if session.chunk_counter == 1:
                     log_info(f"🎤 Started streaming audio to STT", session_id=session.session.session_id)
+                    # İlk chunk'ta format kontrolü yap
+                    if len(decoded_audio) >= 4:
+                        if decoded_audio[:4] == b'\x1a\x45\xdf\xa3':
+                            log_info(f"✅ Valid WEBM header detected", session_id=session.session.session_id)
+                        else:
+                            log_warning(f"⚠️ Unknown audio format, first 4 bytes: {decoded_audio[:4].hex()}", session_id=session.session.session_id)
+                            # Format hatalıysa buffer'ı temizle ve chunk counter'ı resetle
+                            await session.audio_buffer.clear()
+                            session.chunk_counter = 0
+                            return
                 elif session.chunk_counter % 100 == 0:
                     log_info(f"📊 Sent {session.chunk_counter} chunks to STT so far...", session_id=session.session.session_id)
                 
@@ -650,6 +668,8 @@ async def handle_audio_chunk(websocket: WebSocket, session: RealtimeSession, mes
                     log_warning(f"⚠️ STT timeout detected, restarting stream", session_id=session.session.session_id)
                     session.is_streaming = False
                     session.chunk_counter = 0
+                    # Buffer'ı temizle
+                    await session.audio_buffer.clear()
                     # Timeout durumunda yeniden başlat
                     await session.restart_stt_if_needed()
                 else: