Spaces:
Paused
Paused
Create websocket_handler.py
Browse files- websocket_handler.py +427 -0
websocket_handler.py
ADDED
|
@@ -0,0 +1,427 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
WebSocket Handler for Real-time STT/TTS
|
| 3 |
+
"""
|
| 4 |
+
from fastapi import WebSocket, WebSocketDisconnect, HTTPException
|
| 5 |
+
from typing import Dict, Any, Optional
|
| 6 |
+
import json
|
| 7 |
+
import asyncio
|
| 8 |
+
import base64
|
| 9 |
+
from datetime import datetime
|
| 10 |
+
import sys
|
| 11 |
+
import numpy as np
|
| 12 |
+
from enum import Enum
|
| 13 |
+
|
| 14 |
+
from session import Session, session_store
|
| 15 |
+
from config_provider import ConfigProvider
|
| 16 |
+
from chat_handler import _handle_new_message, _handle_parameter_followup
|
| 17 |
+
from stt_factory import STTFactory
|
| 18 |
+
from tts_factory import TTSFactory
|
| 19 |
+
from utils import log
|
| 20 |
+
|
| 21 |
+
# ========================= CONSTANTS =========================
|
| 22 |
+
SILENCE_THRESHOLD_MS = 2000
|
| 23 |
+
AUDIO_CHUNK_SIZE = 4096
|
| 24 |
+
ENERGY_THRESHOLD = 0.01
|
| 25 |
+
|
| 26 |
+
# ========================= ENUMS =========================
|
| 27 |
+
class ConversationState(Enum):
|
| 28 |
+
IDLE = "idle"
|
| 29 |
+
LISTENING = "listening"
|
| 30 |
+
PROCESSING_STT = "processing_stt"
|
| 31 |
+
PROCESSING_LLM = "processing_llm"
|
| 32 |
+
PROCESSING_TTS = "processing_tts"
|
| 33 |
+
PLAYING_AUDIO = "playing_audio"
|
| 34 |
+
|
| 35 |
+
# ========================= CLASSES =========================
|
| 36 |
+
class AudioBuffer:
|
| 37 |
+
"""Buffer for accumulating audio chunks"""
|
| 38 |
+
def __init__(self):
|
| 39 |
+
self.chunks = []
|
| 40 |
+
self.total_size = 0
|
| 41 |
+
|
| 42 |
+
def add_chunk(self, chunk_data: str):
|
| 43 |
+
"""Add base64 encoded audio chunk"""
|
| 44 |
+
decoded = base64.b64decode(chunk_data)
|
| 45 |
+
self.chunks.append(decoded)
|
| 46 |
+
self.total_size += len(decoded)
|
| 47 |
+
|
| 48 |
+
def get_audio(self) -> bytes:
|
| 49 |
+
"""Get concatenated audio data"""
|
| 50 |
+
return b''.join(self.chunks)
|
| 51 |
+
|
| 52 |
+
def clear(self):
|
| 53 |
+
"""Clear buffer"""
|
| 54 |
+
self.chunks.clear()
|
| 55 |
+
self.total_size = 0
|
| 56 |
+
|
| 57 |
+
class SilenceDetector:
|
| 58 |
+
"""Detect silence in audio stream"""
|
| 59 |
+
def __init__(self, threshold_ms: int = SILENCE_THRESHOLD_MS, energy_threshold: float = ENERGY_THRESHOLD):
|
| 60 |
+
self.threshold_ms = threshold_ms
|
| 61 |
+
self.energy_threshold = energy_threshold
|
| 62 |
+
self.silence_start = None
|
| 63 |
+
self.sample_rate = 16000 # Default sample rate
|
| 64 |
+
|
| 65 |
+
def is_silence(self, audio_chunk: bytes) -> bool:
|
| 66 |
+
"""Check if audio chunk is silence"""
|
| 67 |
+
try:
|
| 68 |
+
# Convert bytes to numpy array (assuming 16-bit PCM)
|
| 69 |
+
audio_data = np.frombuffer(audio_chunk, dtype=np.int16)
|
| 70 |
+
|
| 71 |
+
# Calculate RMS energy
|
| 72 |
+
rms = np.sqrt(np.mean(audio_data**2))
|
| 73 |
+
normalized_rms = rms / 32768.0 # Normalize for 16-bit audio
|
| 74 |
+
|
| 75 |
+
return normalized_rms < self.energy_threshold
|
| 76 |
+
except Exception as e:
|
| 77 |
+
log(f"⚠️ Silence detection error: {e}")
|
| 78 |
+
return False
|
| 79 |
+
|
| 80 |
+
def update(self, audio_chunk: bytes) -> Optional[int]:
|
| 81 |
+
"""Update silence detection and return silence duration in ms"""
|
| 82 |
+
is_silent = self.is_silence(audio_chunk)
|
| 83 |
+
|
| 84 |
+
if is_silent:
|
| 85 |
+
if self.silence_start is None:
|
| 86 |
+
self.silence_start = datetime.now()
|
| 87 |
+
log("🔇 Silence started")
|
| 88 |
+
else:
|
| 89 |
+
silence_duration = (datetime.now() - self.silence_start).total_seconds() * 1000
|
| 90 |
+
return int(silence_duration)
|
| 91 |
+
else:
|
| 92 |
+
if self.silence_start is not None:
|
| 93 |
+
log("🔊 Speech detected, silence broken")
|
| 94 |
+
self.silence_start = None
|
| 95 |
+
|
| 96 |
+
return 0
|
| 97 |
+
|
| 98 |
+
class BargeInHandler:
|
| 99 |
+
"""Handle barge-in (interruption) logic"""
|
| 100 |
+
def __init__(self):
|
| 101 |
+
self.interrupted_at_state: Optional[ConversationState] = None
|
| 102 |
+
self.accumulated_text: str = ""
|
| 103 |
+
self.pending_audio_chunks = []
|
| 104 |
+
|
| 105 |
+
def handle_interruption(self, current_state: ConversationState):
|
| 106 |
+
"""Handle user interruption"""
|
| 107 |
+
self.interrupted_at_state = current_state
|
| 108 |
+
log(f"🛑 Barge-in detected at state: {current_state.value}")
|
| 109 |
+
|
| 110 |
+
def should_preserve_context(self) -> bool:
|
| 111 |
+
"""Check if context should be preserved after interruption"""
|
| 112 |
+
# Preserve context if interrupted during LLM or TTS processing
|
| 113 |
+
return self.interrupted_at_state in [
|
| 114 |
+
ConversationState.PROCESSING_LLM,
|
| 115 |
+
ConversationState.PROCESSING_TTS,
|
| 116 |
+
ConversationState.PLAYING_AUDIO
|
| 117 |
+
]
|
| 118 |
+
|
| 119 |
+
class ConversationManager:
|
| 120 |
+
"""Manage conversation state and flow"""
|
| 121 |
+
def __init__(self, session: Session):
|
| 122 |
+
self.session = session
|
| 123 |
+
self.state = ConversationState.IDLE
|
| 124 |
+
self.audio_buffer = AudioBuffer()
|
| 125 |
+
self.silence_detector = SilenceDetector()
|
| 126 |
+
self.barge_in_handler = BargeInHandler()
|
| 127 |
+
self.stt_manager = None
|
| 128 |
+
self.current_transcription = ""
|
| 129 |
+
self.is_streaming = False
|
| 130 |
+
|
| 131 |
+
async def initialize_stt(self):
|
| 132 |
+
"""Initialize STT provider"""
|
| 133 |
+
try:
|
| 134 |
+
self.stt_manager = STTFactory.create_provider()
|
| 135 |
+
if self.stt_manager:
|
| 136 |
+
config = ConfigProvider.get().global_config.stt_settings
|
| 137 |
+
await self.stt_manager.start_streaming({
|
| 138 |
+
"language": config.get("language", "tr-TR"),
|
| 139 |
+
"interim_results": config.get("interim_results", True),
|
| 140 |
+
"single_utterance": False, # Important for continuous listening
|
| 141 |
+
"enable_punctuation": config.get("enable_punctuation", True)
|
| 142 |
+
})
|
| 143 |
+
log("✅ STT manager initialized")
|
| 144 |
+
return True
|
| 145 |
+
except Exception as e:
|
| 146 |
+
log(f"❌ Failed to initialize STT: {e}")
|
| 147 |
+
return False
|
| 148 |
+
|
| 149 |
+
def change_state(self, new_state: ConversationState):
|
| 150 |
+
"""Change conversation state"""
|
| 151 |
+
old_state = self.state
|
| 152 |
+
self.state = new_state
|
| 153 |
+
log(f"📊 State change: {old_state.value} → {new_state.value}")
|
| 154 |
+
|
| 155 |
+
def handle_barge_in(self):
|
| 156 |
+
"""Handle user interruption"""
|
| 157 |
+
self.barge_in_handler.handle_interruption(self.state)
|
| 158 |
+
self.change_state(ConversationState.LISTENING)
|
| 159 |
+
|
| 160 |
+
def reset_audio_buffer(self):
|
| 161 |
+
"""Reset audio buffer for new utterance"""
|
| 162 |
+
self.audio_buffer.clear()
|
| 163 |
+
self.silence_detector.silence_start = None
|
| 164 |
+
self.current_transcription = ""
|
| 165 |
+
|
| 166 |
+
# ========================= WEBSOCKET HANDLER =========================
|
| 167 |
+
async def websocket_endpoint(websocket: WebSocket, session_id: str):
|
| 168 |
+
"""Main WebSocket endpoint for real-time conversation"""
|
| 169 |
+
await websocket.accept()
|
| 170 |
+
log(f"🔌 WebSocket connected for session: {session_id}")
|
| 171 |
+
|
| 172 |
+
# Get session
|
| 173 |
+
session = session_store.get_session(session_id)
|
| 174 |
+
if not session:
|
| 175 |
+
await websocket.send_json({
|
| 176 |
+
"type": "error",
|
| 177 |
+
"message": "Session not found"
|
| 178 |
+
})
|
| 179 |
+
await websocket.close()
|
| 180 |
+
return
|
| 181 |
+
|
| 182 |
+
# Initialize conversation manager
|
| 183 |
+
conversation = ConversationManager(session)
|
| 184 |
+
|
| 185 |
+
# Initialize STT
|
| 186 |
+
stt_initialized = await conversation.initialize_stt()
|
| 187 |
+
if not stt_initialized:
|
| 188 |
+
await websocket.send_json({
|
| 189 |
+
"type": "error",
|
| 190 |
+
"message": "STT initialization failed"
|
| 191 |
+
})
|
| 192 |
+
|
| 193 |
+
try:
|
| 194 |
+
while True:
|
| 195 |
+
# Receive message
|
| 196 |
+
message = await websocket.receive_json()
|
| 197 |
+
message_type = message.get("type")
|
| 198 |
+
|
| 199 |
+
if message_type == "audio_chunk":
|
| 200 |
+
await handle_audio_chunk(websocket, conversation, message)
|
| 201 |
+
|
| 202 |
+
elif message_type == "control":
|
| 203 |
+
await handle_control_message(websocket, conversation, message)
|
| 204 |
+
|
| 205 |
+
elif message_type == "ping":
|
| 206 |
+
# Keep-alive ping
|
| 207 |
+
await websocket.send_json({"type": "pong"})
|
| 208 |
+
|
| 209 |
+
except WebSocketDisconnect:
|
| 210 |
+
log(f"🔌 WebSocket disconnected for session: {session_id}")
|
| 211 |
+
await cleanup_conversation(conversation)
|
| 212 |
+
except Exception as e:
|
| 213 |
+
log(f"❌ WebSocket error: {e}")
|
| 214 |
+
await websocket.send_json({
|
| 215 |
+
"type": "error",
|
| 216 |
+
"message": str(e)
|
| 217 |
+
})
|
| 218 |
+
await cleanup_conversation(conversation)
|
| 219 |
+
|
| 220 |
+
# ========================= MESSAGE HANDLERS =========================
|
| 221 |
+
async def handle_audio_chunk(websocket: WebSocket, conversation: ConversationManager, message: Dict[str, Any]):
|
| 222 |
+
"""Handle incoming audio chunk"""
|
| 223 |
+
try:
|
| 224 |
+
audio_data = message.get("data")
|
| 225 |
+
if not audio_data:
|
| 226 |
+
return
|
| 227 |
+
|
| 228 |
+
# Check for barge-in
|
| 229 |
+
if conversation.state in [ConversationState.PLAYING_AUDIO, ConversationState.PROCESSING_TTS]:
|
| 230 |
+
conversation.handle_barge_in()
|
| 231 |
+
await websocket.send_json({
|
| 232 |
+
"type": "control",
|
| 233 |
+
"action": "stop_playback"
|
| 234 |
+
})
|
| 235 |
+
|
| 236 |
+
# Change state to listening if idle
|
| 237 |
+
if conversation.state == ConversationState.IDLE:
|
| 238 |
+
conversation.change_state(ConversationState.LISTENING)
|
| 239 |
+
await websocket.send_json({
|
| 240 |
+
"type": "state_change",
|
| 241 |
+
"from": "idle",
|
| 242 |
+
"to": "listening"
|
| 243 |
+
})
|
| 244 |
+
|
| 245 |
+
# Add to buffer
|
| 246 |
+
conversation.audio_buffer.add_chunk(audio_data)
|
| 247 |
+
|
| 248 |
+
# Decode for processing
|
| 249 |
+
decoded_audio = base64.b64decode(audio_data)
|
| 250 |
+
|
| 251 |
+
# Check silence
|
| 252 |
+
silence_duration = conversation.silence_detector.update(decoded_audio)
|
| 253 |
+
|
| 254 |
+
# Stream to STT if available
|
| 255 |
+
if conversation.stt_manager and conversation.state == ConversationState.LISTENING:
|
| 256 |
+
async for result in conversation.stt_manager.stream_audio(decoded_audio):
|
| 257 |
+
# Send interim results
|
| 258 |
+
await websocket.send_json({
|
| 259 |
+
"type": "transcription",
|
| 260 |
+
"text": result.text,
|
| 261 |
+
"is_final": result.is_final,
|
| 262 |
+
"confidence": result.confidence
|
| 263 |
+
})
|
| 264 |
+
|
| 265 |
+
if result.is_final:
|
| 266 |
+
conversation.current_transcription = result.text
|
| 267 |
+
|
| 268 |
+
# Check if user stopped speaking (2 seconds of silence)
|
| 269 |
+
if silence_duration > SILENCE_THRESHOLD_MS and conversation.current_transcription:
|
| 270 |
+
log(f"🔇 User stopped speaking after {silence_duration}ms of silence")
|
| 271 |
+
await process_user_input(websocket, conversation)
|
| 272 |
+
|
| 273 |
+
except Exception as e:
|
| 274 |
+
log(f"❌ Audio chunk handling error: {e}")
|
| 275 |
+
await websocket.send_json({
|
| 276 |
+
"type": "error",
|
| 277 |
+
"message": f"Audio processing error: {str(e)}"
|
| 278 |
+
})
|
| 279 |
+
|
| 280 |
+
async def handle_control_message(websocket: WebSocket, conversation: ConversationManager, message: Dict[str, Any]):
|
| 281 |
+
"""Handle control messages"""
|
| 282 |
+
action = message.get("action")
|
| 283 |
+
|
| 284 |
+
if action == "start_session":
|
| 285 |
+
# Session already started
|
| 286 |
+
await websocket.send_json({
|
| 287 |
+
"type": "session_started",
|
| 288 |
+
"session_id": conversation.session.session_id
|
| 289 |
+
})
|
| 290 |
+
|
| 291 |
+
elif action == "end_session":
|
| 292 |
+
# Clean up and close
|
| 293 |
+
await cleanup_conversation(conversation)
|
| 294 |
+
await websocket.close()
|
| 295 |
+
|
| 296 |
+
elif action == "interrupt":
|
| 297 |
+
# Handle explicit interrupt
|
| 298 |
+
conversation.handle_barge_in()
|
| 299 |
+
await websocket.send_json({
|
| 300 |
+
"type": "control",
|
| 301 |
+
"action": "interrupt_acknowledged"
|
| 302 |
+
})
|
| 303 |
+
|
| 304 |
+
elif action == "reset":
|
| 305 |
+
# Reset conversation state
|
| 306 |
+
conversation.reset_audio_buffer()
|
| 307 |
+
conversation.change_state(ConversationState.IDLE)
|
| 308 |
+
await websocket.send_json({
|
| 309 |
+
"type": "state_change",
|
| 310 |
+
"from": conversation.state.value,
|
| 311 |
+
"to": "idle"
|
| 312 |
+
})
|
| 313 |
+
|
| 314 |
+
# ========================= PROCESSING FUNCTIONS =========================
|
| 315 |
+
async def process_user_input(websocket: WebSocket, conversation: ConversationManager):
|
| 316 |
+
"""Process complete user input"""
|
| 317 |
+
try:
|
| 318 |
+
user_text = conversation.current_transcription
|
| 319 |
+
if not user_text:
|
| 320 |
+
conversation.reset_audio_buffer()
|
| 321 |
+
conversation.change_state(ConversationState.IDLE)
|
| 322 |
+
return
|
| 323 |
+
|
| 324 |
+
log(f"💬 Processing user input: {user_text}")
|
| 325 |
+
|
| 326 |
+
# Change state to processing
|
| 327 |
+
conversation.change_state(ConversationState.PROCESSING_STT)
|
| 328 |
+
await websocket.send_json({
|
| 329 |
+
"type": "state_change",
|
| 330 |
+
"from": "listening",
|
| 331 |
+
"to": "processing_stt"
|
| 332 |
+
})
|
| 333 |
+
|
| 334 |
+
# Send final transcription
|
| 335 |
+
await websocket.send_json({
|
| 336 |
+
"type": "transcription",
|
| 337 |
+
"text": user_text,
|
| 338 |
+
"is_final": True,
|
| 339 |
+
"confidence": 0.95
|
| 340 |
+
})
|
| 341 |
+
|
| 342 |
+
# Process with LLM
|
| 343 |
+
conversation.change_state(ConversationState.PROCESSING_LLM)
|
| 344 |
+
await websocket.send_json({
|
| 345 |
+
"type": "state_change",
|
| 346 |
+
"from": "processing_stt",
|
| 347 |
+
"to": "processing_llm"
|
| 348 |
+
})
|
| 349 |
+
|
| 350 |
+
# Add to session history
|
| 351 |
+
conversation.session.add_turn("user", user_text)
|
| 352 |
+
|
| 353 |
+
# Get response based on session state
|
| 354 |
+
if conversation.session.state == "await_param":
|
| 355 |
+
response_text = await _handle_parameter_followup(conversation.session, user_text)
|
| 356 |
+
else:
|
| 357 |
+
response_text = await _handle_new_message(conversation.session, user_text)
|
| 358 |
+
|
| 359 |
+
# Add response to history
|
| 360 |
+
conversation.session.add_turn("assistant", response_text)
|
| 361 |
+
|
| 362 |
+
# Send text response
|
| 363 |
+
await websocket.send_json({
|
| 364 |
+
"type": "assistant_response",
|
| 365 |
+
"text": response_text
|
| 366 |
+
})
|
| 367 |
+
|
| 368 |
+
# Generate TTS if enabled
|
| 369 |
+
tts_provider = TTSFactory.create_provider()
|
| 370 |
+
if tts_provider:
|
| 371 |
+
conversation.change_state(ConversationState.PROCESSING_TTS)
|
| 372 |
+
await websocket.send_json({
|
| 373 |
+
"type": "state_change",
|
| 374 |
+
"from": "processing_llm",
|
| 375 |
+
"to": "processing_tts"
|
| 376 |
+
})
|
| 377 |
+
|
| 378 |
+
# Generate audio
|
| 379 |
+
audio_data = await tts_provider.synthesize(response_text)
|
| 380 |
+
|
| 381 |
+
# Send audio in chunks
|
| 382 |
+
chunk_size = 4096
|
| 383 |
+
for i in range(0, len(audio_data), chunk_size):
|
| 384 |
+
chunk = audio_data[i:i + chunk_size]
|
| 385 |
+
await websocket.send_json({
|
| 386 |
+
"type": "tts_audio",
|
| 387 |
+
"data": base64.b64encode(chunk).decode('utf-8'),
|
| 388 |
+
"chunk_index": i // chunk_size,
|
| 389 |
+
"is_last": i + chunk_size >= len(audio_data)
|
| 390 |
+
})
|
| 391 |
+
|
| 392 |
+
conversation.change_state(ConversationState.PLAYING_AUDIO)
|
| 393 |
+
await websocket.send_json({
|
| 394 |
+
"type": "state_change",
|
| 395 |
+
"from": "processing_tts",
|
| 396 |
+
"to": "playing_audio"
|
| 397 |
+
})
|
| 398 |
+
else:
|
| 399 |
+
# No TTS, go back to idle
|
| 400 |
+
conversation.change_state(ConversationState.IDLE)
|
| 401 |
+
await websocket.send_json({
|
| 402 |
+
"type": "state_change",
|
| 403 |
+
"from": "processing_llm",
|
| 404 |
+
"to": "idle"
|
| 405 |
+
})
|
| 406 |
+
|
| 407 |
+
# Reset for next input
|
| 408 |
+
conversation.reset_audio_buffer()
|
| 409 |
+
|
| 410 |
+
except Exception as e:
|
| 411 |
+
log(f"❌ Error processing user input: {e}")
|
| 412 |
+
await websocket.send_json({
|
| 413 |
+
"type": "error",
|
| 414 |
+
"message": f"Processing error: {str(e)}"
|
| 415 |
+
})
|
| 416 |
+
conversation.reset_audio_buffer()
|
| 417 |
+
conversation.change_state(ConversationState.IDLE)
|
| 418 |
+
|
| 419 |
+
# ========================= CLEANUP =========================
|
| 420 |
+
async def cleanup_conversation(conversation: ConversationManager):
|
| 421 |
+
"""Clean up conversation resources"""
|
| 422 |
+
try:
|
| 423 |
+
if conversation.stt_manager:
|
| 424 |
+
await conversation.stt_manager.stop_streaming()
|
| 425 |
+
log(f"🧹 Cleaned up conversation for session: {conversation.session.session_id}")
|
| 426 |
+
except Exception as e:
|
| 427 |
+
log(f"⚠️ Cleanup error: {e}")
|