websocket_handler.py

"""
WebSocket Handler for Real-time STT/TTS with Barge-in Support
"""
from fastapi import WebSocket, WebSocketDisconnect
from typing import Dict, Any, Optional
import json
import asyncio
import base64
from datetime import datetime
from collections import deque
from enum import Enum
import numpy as np
import traceback

from session import Session, session_store
from config_provider import ConfigProvider
from chat_handler import handle_new_message, handle_parameter_followup
from stt_factory import STTFactory
from tts_factory import TTSFactory
from logger import log_info, log_error, log_debug, log_warning

# ========================= CONSTANTS =========================
# Default values - will be overridden by config
DEFAULT_SILENCE_THRESHOLD_MS = 2000
DEFAULT_AUDIO_CHUNK_SIZE = 4096
DEFAULT_ENERGY_THRESHOLD = 0.0005 # 0.01
DEFAULT_AUDIO_BUFFER_MAX_SIZE = 1000

# ========================= ENUMS =========================
class ConversationState(Enum):
    IDLE = "idle"
    LISTENING = "listening"
    PROCESSING_STT = "processing_stt"
    PROCESSING_LLM = "processing_llm"
    PROCESSING_TTS = "processing_tts"
    PLAYING_AUDIO = "playing_audio"

# ========================= CLASSES =========================
class AudioBuffer:
    """Thread-safe circular buffer for audio chunks"""
    def __init__(self, max_size: int = DEFAULT_AUDIO_BUFFER_MAX_SIZE):
        self.buffer = deque(maxlen=max_size)
        self.lock = asyncio.Lock()
        
    async def add_chunk(self, chunk_data: str):
        """Add base64 encoded audio chunk"""
        async with self.lock:
            decoded = base64.b64decode(chunk_data)
            self.buffer.append(decoded)
            
    async def get_all_audio(self) -> bytes:
        """Get all audio data concatenated"""
        async with self.lock:
            return b''.join(self.buffer)
    
    async def clear(self):
        """Clear buffer"""
        async with self.lock:
            self.buffer.clear()
    
    def size(self) -> int:
        """Get current buffer size"""
        return len(self.buffer)


class SilenceDetector:
    """Detect silence in audio stream"""
    def __init__(self, threshold_ms: int = DEFAULT_SILENCE_THRESHOLD_MS, energy_threshold: float = DEFAULT_ENERGY_THRESHOLD):
        self.threshold_ms = threshold_ms
        self.energy_threshold = energy_threshold
        self.silence_start = None
        self.sample_rate = 16000
        
    def update(self, audio_chunk: bytes) -> int:
        """Update with new audio chunk and return silence duration in ms"""
        if self.is_silence(audio_chunk):
            if self.silence_start is None:
                self.silence_start = datetime.now()
            silence_duration = (datetime.now() - self.silence_start).total_seconds() * 1000
            return int(silence_duration)
        else:
            self.silence_start = None
            return 0
    
    def is_silence(self, audio_chunk: bytes) -> bool:
        """Check if audio chunk is silence"""
        try:
            # Audio chunk boyutunu kontrol et
            if len(audio_chunk) == 0:
                return True
            
            # Chunk boyutu 2'nin katı olmalı (16-bit audio için)
            if len(audio_chunk) % 2 != 0:
                # Tek byte varsa, son byte'ı at
                audio_chunk = audio_chunk[:-1]
                
            # Convert bytes to numpy array (assuming 16-bit PCM)
            audio_data = np.frombuffer(audio_chunk, dtype=np.int16)
            
            # RMS hesapla
            if len(audio_data) == 0:
                return True
                
            rms = np.sqrt(np.mean(audio_data.astype(float) ** 2))
            normalized_rms = rms / 32768.0
            
            return normalized_rms < self.energy_threshold
            
        except Exception as e:
            log_warning(f"Silence detection error: {e}")
            return False
    
    def reset(self):
        """Reset silence detection"""
        self.silence_start = None


class BargeInHandler:
    """Handle user interruptions during TTS playback"""
    def __init__(self):
        self.active_tts_task: Optional[asyncio.Task] = None
        self.is_interrupting = False
        self.lock = asyncio.Lock()
    
    async def start_tts_task(self, coro):
        """Start a cancellable TTS task"""
        async with self.lock:
            # Cancel any existing task
            if self.active_tts_task and not self.active_tts_task.done():
                self.active_tts_task.cancel()
                try:
                    await self.active_tts_task
                except asyncio.CancelledError:
                    pass
            
            # Start new task
            self.active_tts_task = asyncio.create_task(coro)
            return self.active_tts_task
    
    async def handle_interruption(self, current_state: ConversationState):
        """Handle barge-in interruption"""
        async with self.lock:
            self.is_interrupting = True
            
            # Cancel TTS if active
            if self.active_tts_task and not self.active_tts_task.done():
                log_info("Barge-in: Cancelling active TTS")
                self.active_tts_task.cancel()
                try:
                    await self.active_tts_task
                except asyncio.CancelledError:
                    pass
            
            # Reset flag after short delay
            await asyncio.sleep(0.5)
            self.is_interrupting = False


class RealtimeSession:
    """Manage a real-time conversation session"""
    def __init__(self, session: Session):
        self.session = session
        self.state = ConversationState.IDLE
        self.is_websocket_active = True
        
        # Get settings from config
        config = ConfigProvider.get().global_config.stt_provider.settings
        
        # Initialize with config values or defaults
        silence_threshold = config.get("speech_timeout_ms", DEFAULT_SILENCE_THRESHOLD_MS)
        energy_threshold = config.get("energy_threshold", DEFAULT_ENERGY_THRESHOLD)
        buffer_max_size = config.get("audio_buffer_max_size", DEFAULT_AUDIO_BUFFER_MAX_SIZE)
        
        self.audio_buffer = AudioBuffer(max_size=buffer_max_size)
        self.silence_detector = SilenceDetector(
            threshold_ms=silence_threshold,
            energy_threshold=energy_threshold
        )
        self.barge_in_handler = BargeInHandler()
        self.stt_manager = None
        self.current_transcription = ""
        self.is_streaming = False
        self.lock = asyncio.Lock()
        
        # Store config for later use
        self.audio_chunk_size = config.get("audio_chunk_size", DEFAULT_AUDIO_CHUNK_SIZE)
        self.silence_threshold_ms = silence_threshold
        
        # Chunk counter için attribute
        self.chunk_counter = 0
        
        # Session management - YENİ
        self.stt_session_count = 0
        self.last_stt_stop_time = None
        
    async def initialize_stt(self):
        """Initialize STT provider with clean state"""
        try:
            # Session numarasını artır
            self.stt_session_count += 1
            log_info(f"🎤 Initializing STT session #{self.stt_session_count}", session_id=self.session.session_id)
            
            # Önce mevcut STT'yi tamamen temizle
            await self.stop_stt_streaming()
            
            # Önceki stop'tan bu yana yeterli zaman geçtiğinden emin ol
            if self.last_stt_stop_time:
                elapsed = (datetime.now() - self.last_stt_stop_time).total_seconds()
                if elapsed < 0.5:
                    wait_time = 0.5 - elapsed
                    log_info(f"⏳ Waiting {wait_time:.2f}s for proper cleanup", session_id=self.session.session_id)
                    await asyncio.sleep(wait_time)
            
            # Tüm değişkenleri yeniden başlat
            self.chunk_counter = 0
            self.current_transcription = ""
            await self.audio_buffer.clear()
            self.silence_detector.reset()
            
            # Yeni STT instance oluştur
            self.stt_manager = STTFactory.create_provider()
            if not self.stt_manager:
                log_error("❌ STT manager is None - STTFactory.create_provider() returned None", session_id=self.session.session_id)
                return False
                
            log_info(f"✅ STT manager created: {type(self.stt_manager).__name__}", session_id=self.session.session_id)
            
            # Get STT config from provider settings
            config = ConfigProvider.get().global_config.stt_provider.settings
            
            # Get language from session locale
            session_locale = getattr(self.session, 'locale', 'tr')
            
            # Import LocaleManager to get proper locale tag
            from locale_manager import LocaleManager
            locale_data = LocaleManager.get_locale(session_locale)
            
            # Get proper locale tag for STT (e.g., tr -> tr-TR)
            language_code = locale_data.get('locale_tag', 'tr-TR')
            
            log_info(f"🌍 Session locale: {session_locale}, STT language: {language_code}", session_id=self.session.session_id)
            
            # single_utterance'ı false yap - sürekli dinleme için
            stt_config = {
                "language": language_code,
                "interim_results": config.get("interim_results", True),
                "single_utterance": False,  # Sürekli dinleme için false
                "enable_punctuation": config.get("enable_punctuation", True),
                "sample_rate": 16000,
                "encoding": "WEBM_OPUS"
            }
            
            log_info(f"🎤 Starting STT streaming with config: {stt_config}", session_id=self.session.session_id)
            
            # Start streaming
            await self.stt_manager.start_streaming(stt_config)
            self.is_streaming = True
            
            log_info("✅ STT streaming started successfully with clean state", session_id=self.session.session_id)
            return True
            
        except Exception as e:
            log_error(f"❌ Failed to initialize STT", error=str(e), traceback=traceback.format_exc(), session_id=self.session.session_id)
            # Hata durumunda da temizlik yap
            await self.stop_stt_streaming()
            return False

    async def stop_stt_streaming(self):
        """Stop STT streaming completely and reset all STT-related variables"""
        try:
            log_info(f"🛑 Stopping STT session #{self.stt_session_count}", session_id=self.session.session_id)
            
            # STT manager varsa durdur
            if self.stt_manager:
                if self.is_streaming:
                    try:
                        await self.stt_manager.stop_streaming()
                    except Exception as e:
                        log_warning(f"⚠️ Error during STT stop_streaming: {e}", session_id=self.session.session_id)
                
                # STT manager'ı tamamen sil
                self.stt_manager = None
            
            # Tüm STT ile ilgili değişkenleri resetle
            self.is_streaming = False
            self.chunk_counter = 0
            self.current_transcription = ""
            
            # Audio buffer'ı temizle
            await self.audio_buffer.clear()
            
            # Silence detector'ı resetle
            self.silence_detector.reset()
            
            # Speech started flag'ini temizle
            if hasattr(self, 'speech_started'):
                delattr(self, 'speech_started')
            
            # Stop zamanını kaydet
            self.last_stt_stop_time = datetime.now()
            
            log_info(f"✅ STT session #{self.stt_session_count} stopped and all data reset", session_id=self.session.session_id)
            
        except Exception as e:
            log_error(f"❌ Error in stop_stt_streaming", error=str(e), session_id=self.session.session_id)
            # Hata olsa bile değişkenleri resetle
            self.stt_manager = None
            self.is_streaming = False
            self.chunk_counter = 0
            self.current_transcription = ""
            if self.audio_buffer:
                await self.audio_buffer.clear()
            if self.silence_detector:
                self.silence_detector.reset()
            self.last_stt_stop_time = datetime.now()

    async def restart_stt_if_needed(self):
        """Restart STT if it's not active"""
        try:
            # Sadece LISTENING state'inde ve WebSocket aktifse restart yap
            if not self.is_streaming and self.is_websocket_active and self.state == ConversationState.LISTENING:
                log_info(f"🔄 Restarting STT stream (session #{self.stt_session_count} -> #{self.stt_session_count + 1})", 
                        session_id=self.session.session_id)
                
                # Yeni session başlat (initialize_stt zaten stop_stt_streaming'i çağırıyor)
                stt_initialized = await self.initialize_stt()
                if stt_initialized:
                    log_info(f"✅ STT stream restarted successfully", session_id=self.session.session_id)
                    return True
                else:
                    log_error(f"❌ Failed to restart STT stream", session_id=self.session.session_id)
                    return False
            return True
        except Exception as e:
            log_error(f"❌ Error restarting STT", error=str(e), traceback=traceback.format_exc(), 
                     session_id=self.session.session_id)
            return False
    
    async def change_state(self, new_state: ConversationState):
        """Change conversation state"""
        async with self.lock:
            old_state = self.state
            self.state = new_state
            log_debug(
                f"State change: {old_state.value} → {new_state.value}",
                session_id=self.session.session_id
            )
    
    async def handle_barge_in(self):
        """Handle user interruption"""
        # Barge-in devre dışı - bu metod artık çağrılmamalı
        log_warning(f"⚠️ Barge-in called but disabled", session_id=self.session.session_id)
        return
    
    async def reset_for_new_utterance(self):
        """Reset for new user utterance"""
        log_info(f"🔄 Resetting for new utterance", session_id=self.session.session_id)
        
        # Buffer ve detector'ı temizle
        await self.audio_buffer.clear()
        self.silence_detector.reset()
        
        # Transcription ve counter'ı sıfırla
        self.current_transcription = ""
        self.chunk_counter = 0
        
        # Speech started flag'ini temizle
        if hasattr(self, 'speech_started'):
            delattr(self, 'speech_started')
            
        log_info(f"✅ Reset for new utterance complete", session_id=self.session.session_id)
    
    async def cleanup(self):
        """Clean up resources"""
        try:
            self.is_websocket_active = False
            await self.stop_stt_streaming()  # STT'yi düzgün durdur
            log_info(f"Cleaned up realtime session", session_id=self.session.session_id)
        except Exception as e:
            log_warning(f"Cleanup error", error=str(e), session_id=self.session.session_id)

# ========================= MAIN HANDLER =========================
async def websocket_endpoint(websocket: WebSocket, session_id: str):
    """Main WebSocket endpoint for real-time conversation"""
    log_info(f"🔌 WebSocket connection attempt", session_id=session_id)
    
    await websocket.accept()
    log_info(f"✅ WebSocket accepted", session_id=session_id)
    
    # Get session
    session = session_store.get_session(session_id)
    if not session:
        log_error(f"❌ Session not found", session_id=session_id)
        await websocket.send_json({
            "type": "error",
            "message": "Session not found"
        })
        await websocket.close()
        return
    
    log_info(f"✅ Session found", session_id=session_id, project=session.project_name)
    
    # Mark as realtime session
    session.is_realtime = True
    session_store.update_session(session)
    
    # Initialize conversation
    realtime_session = RealtimeSession(session)
    
    # Initialize STT
    log_info(f"🎤 Initializing STT...", session_id=session_id)
    stt_initialized = await realtime_session.initialize_stt()
    if not stt_initialized:
        log_error(f"❌ STT initialization failed", session_id=session_id)
        await websocket.send_json({
            "type": "error",
            "message": "STT initialization failed"
        })
    else:
        log_info(f"✅ STT initialized", session_id=session_id)
    
    # Send session started confirmation
    await websocket.send_json({
        "type": "session_started",
        "session_id": session_id,
        "stt_initialized": stt_initialized
    })
    
    # Send welcome message from session history
    log_info(f"📋 Checking for welcome message in session history...", session_id=session_id)
    
    # chat_history değişkenini session'dan al
    chat_history = session.chat_history
    
    if chat_history and len(chat_history) > 0:
        log_info(f"📋 Found {len(chat_history)} messages in history", session_id=session_id)
        
        # Get the last assistant message (welcome message)
        for i, msg in enumerate(reversed(chat_history)):
            log_debug(f"📋 Message {i}: role={msg.get('role', 'unknown')}, content_preview={msg.get('content', '')[:50]}...", session_id=session_id)
            
            if msg.get('role') == 'assistant':
                welcome_text = msg.get('content', '')
                log_info(f"📢 Found welcome message: {welcome_text[:50]}...", session_id=session_id)
                
                # Send text first
                try:
                    await websocket.send_json({
                        "type": "assistant_response",
                        "text": welcome_text,
                        "is_welcome": True
                    })
                    log_info(f"✅ Welcome text sent via WebSocket", session_id=session_id)
                except Exception as e:
                    log_error(f"❌ Failed to send welcome text", error=str(e), session_id=session_id)
                
                # Generate and send TTS if available
                tts_provider = TTSFactory.create_provider()
                if tts_provider:
                    try:
                        log_info(f"🎤 Generating welcome TTS...", session_id=session_id)
                        
                        # TTS preprocessor kullan
                        from tts_preprocessor import TTSPreprocessor
                        preprocessor = TTSPreprocessor(language=session.locale)
                        processed_text = preprocessor.preprocess(
                            welcome_text,
                            tts_provider.get_preprocessing_flags()
                        )
                        
                        # TTS oluştur
                        audio_data = await tts_provider.synthesize(processed_text)
                        
                        if audio_data:
                            # Audio'yu base64'e çevir ve chunk'lara böl
                            audio_base64 = base64.b64encode(audio_data).decode('utf-8')
                            chunk_size = 16384
                            total_length = len(audio_base64)
                            total_chunks = (total_length + chunk_size - 1) // chunk_size
                            
                            log_info(f"📤 Sending welcome TTS in {total_chunks} chunks", session_id=session_id)
                            
                            for i in range(0, total_length, chunk_size):
                                chunk = audio_base64[i:i + chunk_size]
                                chunk_index = i // chunk_size
                                is_last = chunk_index == total_chunks - 1
                                
                                await websocket.send_json({
                                    "type": "tts_audio",
                                    "data": chunk,
                                    "chunk_index": chunk_index,
                                    "total_chunks": total_chunks,
                                    "is_last": is_last,
                                    "mime_type": "audio/mpeg"
                                })
                            
                            log_info(f"✅ Welcome TTS sent", session_id=session_id)
                    except Exception as e:
                        log_error(f"❌ Failed to send welcome TTS", error=str(e), traceback=traceback.format_exc(), session_id=session_id)
                else:
                    log_warning(f"⚠️ No TTS provider available", session_id=session_id)
                
                break
        else:
            log_warning(f"⚠️ No assistant message found in history", session_id=session_id)
    else:
        log_warning(f"⚠️ No messages in session history", session_id=session_id)
    
    log_info(f"💬 Ready for conversation", session_id=session_id)

    try:
        while True:
            try:
                # WebSocket aktif mi kontrol et
                if not realtime_session.is_websocket_active:
                    log_info(f"🔌 WebSocket inactive, breaking loop", session_id=session_id)
                    break
                    
                # Receive message with timeout
                message = await asyncio.wait_for(
                    websocket.receive_json(), 
                    timeout=60.0  # 60 second timeout
                )
                
                message_type = message.get("type")
                # Debug log'u kaldırdık
                
                if message_type == "audio_chunk":
                    await handle_audio_chunk(websocket, realtime_session, message)
                    
                elif message_type == "control":
                    await handle_control_message(websocket, realtime_session, message)
                    
                elif message_type == "ping":
                    # Keep-alive ping - log yapmadan
                    if realtime_session.is_websocket_active:
                        await websocket.send_json({"type": "pong"})
                    
            except asyncio.TimeoutError:
                # Timeout log'unu da azaltalım - her timeout'ta değil
                if realtime_session.is_websocket_active:
                    await websocket.send_json({"type": "ping"})
                    
    except WebSocketDisconnect as e:
        log_info(f"🔌 WebSocket disconnected", session_id=session_id, code=e.code, reason=e.reason)
    except Exception as e:
        # WebSocket kapalıysa hata verme
        if "WebSocket is not connected" not in str(e) and "Cannot call \"send\"" not in str(e):
            log_error(
                f"❌ WebSocket error",
                error=str(e),
                traceback=traceback.format_exc(),
                session_id=session_id
            )
        
        # Error mesajı göndermeye çalışma, zaten kapalı olabilir
        if realtime_session.is_websocket_active:
            try:
                await websocket.send_json({
                    "type": "error",
                    "message": str(e)
                })
            except:
                pass
    finally:
        log_info(f"🧹 Cleaning up WebSocket connection", session_id=session_id)
        await realtime_session.cleanup()
        
        # WebSocket'in açık olup olmadığını kontrol et
        try:
            if websocket.client_state.value == 1:  # 1 = CONNECTED state
                await websocket.close()
        except Exception as e:
            log_debug(f"WebSocket already closed or error during close: {e}", session_id=session_id)
        
# ========================= MESSAGE HANDLERS =========================
async def handle_audio_chunk(websocket: WebSocket, session: RealtimeSession, message: Dict[str, Any]):
    """Handle incoming audio chunk with sequential processing"""
    try:
        # WebSocket kapandıysa işlem yapma
        if not session.is_websocket_active:
            return
            
        audio_data = message.get("data")
        if not audio_data:
            log_warning(f"⚠️ Empty audio chunk received", session_id=session.session.session_id)
            return
        
        # Barge-in devre dışı - TTS/audio playback sırasında audio chunk'ları işleme
        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",
                "from": "idle",
                "to": "listening"
            })
            # IDLE'dan LISTENING'e geçerken STT'yi başlat
            if not session.is_streaming:
                await session.restart_stt_if_needed()
        
        # LISTENING state'inde değilse audio işleme
        if session.state != ConversationState.LISTENING:
            return
            
        # Add to buffer
        await session.audio_buffer.add_chunk(audio_data)
        
        # Decode for processing
        decoded_audio = base64.b64decode(audio_data)
        
        # Check silence
        silence_duration = session.silence_detector.update(decoded_audio)
        
        # Stream to STT if available and in LISTENING state
        if session.stt_manager and session.state == ConversationState.LISTENING:
            # Ensure streaming is active
            if not session.is_streaming:
                log_warning(f"⚠️ STT not streaming, attempting to restart", session_id=session.session.session_id)
                restart_success = await session.restart_stt_if_needed()
                if not restart_success:
                    await websocket.send_json({
                        "type": "error",
                        "error_type": "stt_error",
                        "message": "STT streaming not available"
                    })
                    return
            
            try:
                # Chunk counter artır
                session.chunk_counter += 1
                
                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)
                
                # STT'ye gönder ve sonuçları bekle
                async for result in session.stt_manager.stream_audio(decoded_audio):
                    # SADECE FINAL RESULT'LARI İŞLE
                    if result.is_final:
                        log_info(f"✅ FINAL TRANSCRIPTION: '{result.text}'", session_id=session.session.session_id)
                        
                        # Send ONLY final transcription to frontend
                        await websocket.send_json({
                            "type": "transcription",
                            "text": result.text,
                            "is_final": True,
                            "confidence": result.confidence
                        })
                        
                        session.current_transcription = result.text
                        
                        # Final transcription geldiğinde STT'yi durdur ve işle
                        if session.current_transcription:
                            # Önce STT'yi durdur
                            await session.stop_stt_streaming()
                            
                            # State'i değiştir
                            await session.change_state(ConversationState.PROCESSING_STT)
                            await websocket.send_json({
                                "type": "state_change",
                                "from": "listening",
                                "to": "processing_stt"
                            })
                            
                            # Process user input
                            await process_user_input(websocket, session)
                            
                            # Reset for new utterance
                            await session.reset_for_new_utterance()
                            return
                        
            except Exception as e:
                error_msg = str(e)
                # Google STT timeout hatası kontrolü
                if "Audio Timeout Error" in error_msg or "stream duration" in error_msg or "Exceeded maximum allowed stream duration" in error_msg:
                    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:
                    log_error(f"❌ STT streaming error", error=error_msg, traceback=traceback.format_exc(), session_id=session.session.session_id)
                    await websocket.send_json({
                        "type": "error",
                        "error_type": "stt_error",
                        "message": f"STT error: {str(e)}"
                    })
            
    except Exception as e:
        log_error(f"❌ Error in handle_audio_chunk", error=str(e), traceback=traceback.format_exc(), session_id=session.session.session_id)
        await websocket.send_json({
            "type": "error",
            "error_type": "audio_error",
            "message": f"Audio processing error: {str(e)}"
        })
        
async def handle_control_message(websocket: WebSocket, session: RealtimeSession, message: Dict[str, Any]):
    """Handle control messages"""
    action = message.get("action")
    config = message.get("config", {})
    
    log_debug(f"🎮 Control message", action=action, session_id=session.session.session_id)
    
    if action == "start_session":
        # Session configuration
        await websocket.send_json({
            "type": "session_config",
            "session_id": session.session.session_id,
            "config": {
                "silence_threshold_ms": session.silence_threshold_ms,
                "audio_chunk_size": session.audio_chunk_size,
                "supports_barge_in": False  # Barge-in devre dışı
            }
        })
        
    elif action == "end_session" or action == "stop_session":
        # Clean up and close
        await session.cleanup()
        await websocket.close()
        
    elif action == "interrupt":
        # Barge-in devre dışı - ignore
        log_warning(f"⚠️ Interrupt request ignored (barge-in disabled)", session_id=session.session.session_id)
        
    elif action == "reset":
        # Reset conversation state
        await session.reset_for_new_utterance()
        await session.stop_stt_streaming()
        await session.change_state(ConversationState.IDLE)
        await websocket.send_json({
            "type": "state_change",
            "from": session.state.value,
            "to": "idle"
        })
        
    elif action == "audio_ended":
        # Audio playback ended on client
        if session.state == ConversationState.PLAYING_AUDIO:
            log_info(f"🎵 Client reported audio ended", session_id=session.session.session_id)
            await session.change_state(ConversationState.LISTENING)
            await websocket.send_json({
                "type": "state_change",
                "from": "playing_audio",
                "to": "listening"
            })
            # STT'yi yeniden başlat
            await session.restart_stt_if_needed()
            
    elif action == "restart_stt":
        # Manual STT restart request
        log_info(f"🔄 Manual STT restart requested", session_id=session.session.session_id)
        await session.stop_stt_streaming()
        await session.restart_stt_if_needed()

# ========================= PROCESSING FUNCTIONS =========================
async def process_user_input(websocket: WebSocket, session: RealtimeSession):
    """Process complete user input"""
    try:
        # LLM işlemesi başlamadan önce STT'nin tamamen durduğundan emin ol
        await session.stop_stt_streaming()

        # WebSocket aktif mi kontrol et
        if not session.is_websocket_active:
            return
            
        user_text = session.current_transcription
        if not user_text:
            log_warning(f"⚠️ Empty transcription, returning to listening", session_id=session.session.session_id)
            # Boş transcription durumunda listening'e dön ve STT'yi yeniden başlat
            await session.change_state(ConversationState.LISTENING)
            await session.restart_stt_if_needed()
            return
        
        log_info(f"🎯 Processing user input", text=user_text, session_id=session.session.session_id)
        
        # Send final transcription
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "transcription",
                "text": user_text,
                "is_final": True,
                "confidence": 0.95
            })
        
        # State: LLM Processing
        await session.change_state(ConversationState.PROCESSING_LLM)
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "state_change",
                "from": "processing_stt",
                "to": "processing_llm"
            })
        
        # Add to chat history
        session.session.add_message("user", user_text)
        
        # Get LLM response based on session state
        log_info(f"🤖 Getting LLM response", session_state=session.session.state, session_id=session.session.session_id)
        
        if session.session.state == "collect_params":
            response_text = await handle_parameter_followup(session.session, user_text)
        else:
            response_text = await handle_new_message(session.session, user_text)
        
        log_info(f"💬 LLM response: {response_text[:50]}...", session_id=session.session.session_id)
        
        # Add response to history
        session.session.add_message("assistant", response_text)
        
        # Send text response
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "assistant_response",
                "text": response_text
            })
        
        # Generate TTS if enabled
        tts_provider = TTSFactory.create_provider()
        log_info(f"🔍 TTS provider check: {tts_provider is not None}", session_id=session.session.session_id)
        
        if tts_provider and session.is_websocket_active:
            await session.change_state(ConversationState.PROCESSING_TTS)
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "state_change",
                    "from": "processing_llm",
                    "to": "processing_tts"
                })
            
            log_info(f"🎵 Starting TTS generation for response", session_id=session.session.session_id)
            
            # Generate TTS (barge-in devre dışı)
            await generate_and_stream_tts(websocket, session, tts_provider, response_text)
            
            # TTS bittikten sonra LISTENING state'ine geç
            await session.change_state(ConversationState.LISTENING)
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "state_change",
                    "from": "playing_audio",
                    "to": "listening"
                })
            
            # STT'yi yeniden başlat
            log_info(f"🔄 Restarting STT after TTS completion", session_id=session.session.session_id)
            await session.restart_stt_if_needed()
            
        else:
            log_info(f"⚠️ No TTS provider or WebSocket inactive", session_id=session.session.session_id)
            # No TTS, go back to listening and restart STT
            await session.change_state(ConversationState.LISTENING)
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "state_change",
                    "from": "processing_llm",
                    "to": "listening"
                })
            await session.restart_stt_if_needed()
        
    except Exception as e:
        log_error(
            f"❌ Error processing user input",
            error=str(e),
            traceback=traceback.format_exc(),
            session_id=session.session.session_id
        )
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "error",
                "message": f"Processing error: {str(e)}"
            })
        await session.reset_for_new_utterance()
        # Hata durumunda listening'e dön ve STT'yi yeniden başlat
        await session.change_state(ConversationState.LISTENING)
        await session.restart_stt_if_needed()

async def generate_and_stream_tts(
    websocket: WebSocket,
    session: RealtimeSession,
    tts_provider,
    text: str
):
    """Generate and stream TTS audio with sequential processing"""
    try:
        # TTS başlamadan önce STT'nin tamamen durduğundan emin ol
        await session.stop_stt_streaming()
        
        log_info(f"🎤 Starting TTS generation for text: '{text[:50]}...'", session_id=session.session.session_id)
        
        # TTS preprocessor kullan
        from tts_preprocessor import TTSPreprocessor
        preprocessor = TTSPreprocessor(language=session.session.locale)
        processed_text = preprocessor.preprocess(
            text,
            tts_provider.get_preprocessing_flags()
        )
        
        log_debug(f"📝 Preprocessed text: '{processed_text[:50]}...'", session_id=session.session.session_id)
        
        # Generate audio
        audio_data = await tts_provider.synthesize(processed_text)
        log_info(f"✅ TTS generated: {len(audio_data)} bytes, type: {type(audio_data)}", session_id=session.session.session_id)
        
        # WebSocket aktif mi kontrol et
        if not session.is_websocket_active:
            log_warning(f"⚠️ WebSocket inactive, skipping TTS streaming", session_id=session.session.session_id)
            return
        
        # Change state to playing
        await session.change_state(ConversationState.PLAYING_AUDIO)
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "state_change",
                "from": "processing_tts",
                "to": "playing_audio"
            })
        
        # Convert entire audio to base64 for transmission
        import base64
        log_debug(f"📦 Converting audio to base64...")
        audio_base64 = base64.b64encode(audio_data).decode('utf-8')
        log_info(f"📊 Base64 conversion complete: {len(audio_base64)} chars from {len(audio_data)} bytes", session_id=session.session.session_id)
        
        # Log first 100 chars of base64 to verify it's valid
        log_debug(f"🔍 Base64 preview: {audio_base64[:100]}...")
        
        # Stream audio in chunks
        chunk_size = 16384  # Larger chunk size for base64
        total_length = len(audio_base64)
        total_chunks = (total_length + chunk_size - 1) // chunk_size
        
        log_info(f"📤 Streaming TTS audio: {len(audio_data)} bytes as {total_length} base64 chars in {total_chunks} chunks", session_id=session.session.session_id)
        
        for i in range(0, total_length, chunk_size):
            # WebSocket aktif mi kontrol et
            if not session.is_websocket_active:
                log_warning(f"⚠️ WebSocket inactive during streaming, stopping", session_id=session.session.session_id)
                break
                
            chunk = audio_base64[i:i + chunk_size]
            chunk_index = i // chunk_size
            is_last = chunk_index == total_chunks - 1
            
            log_debug(f"📨 Sending chunk {chunk_index}/{total_chunks}, size: {len(chunk)}, is_last: {is_last}")
            
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "tts_audio",
                    "data": chunk,
                    "chunk_index": chunk_index,
                    "total_chunks": total_chunks,
                    "is_last": is_last,
                    "mime_type": "audio/mpeg"
                })
            
            # Small delay to prevent overwhelming the client
            await asyncio.sleep(0.01)
        
        log_info(
            f"✅ TTS streaming completed successfully",
            session_id=session.session.session_id,
            text_length=len(text),
            audio_size=len(audio_data),
            chunks_sent=total_chunks
        )
        
        # TTS bitimi - state değişimi process_user_input'ta yapılacak
        
    except Exception as e:
        error_msg = str(e)
        log_error(
            f"❌ TTS generation error",
            error=error_msg,
            traceback=traceback.format_exc(),
            session_id=session.session.session_id
        )
        
        # Quota hatası için özel handling
        if "quota_exceeded" in error_msg:
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "tts_error",
                    "message": "TTS servisinin kredi limiti aşıldı. Yanıt sadece metin olarak gösterilecek.",
                    "error_type": "quota_exceeded"
                })
        else:
            if session.is_websocket_active:
                await websocket.send_json({
                    "type": "error",
                    "message": f"TTS error: {error_msg}"
                })
        
        # TTS hatası durumunda listening'e dön
        await session.change_state(ConversationState.LISTENING)
        if session.is_websocket_active:
            await websocket.send_json({
                "type": "state_change", 
                "from": "processing_tts",
                "to": "listening"
            })
        # STT'yi yeniden başlat
        await session.restart_stt_if_needed()