237 lines
9.3 KiB
Python
237 lines
9.3 KiB
Python
import asyncio
|
|
from services.audio_session import AudioSession
|
|
from services.task_manager import task_manager
|
|
from utils.logger import session_logger
|
|
from services.interrupt_handler import interrupt_handler
|
|
import time
|
|
from config import settings
|
|
|
|
|
|
async def _connect_with_interrupt_cleanup(service, session_key):
|
|
"""Own warmup connections even when interruption races with connect()."""
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
return False
|
|
await interrupt_handler.register_cleanup_handler(
|
|
session_key, service.close,
|
|
priority=interrupt_handler.PRIORITY["NETWORK_CONNECTION"],
|
|
)
|
|
try:
|
|
await service.connect()
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
await service.close()
|
|
return False
|
|
return True
|
|
except asyncio.CancelledError:
|
|
await service.close()
|
|
raise
|
|
except Exception:
|
|
await service.close()
|
|
raise
|
|
|
|
|
|
async def prepare_asr_service(session, device_id, session_id):
|
|
session.asr_error = None
|
|
session.asr_service.sample_rate = session.sample_rate
|
|
if hasattr(session.asr_service, "device_id"):
|
|
session.asr_service.device_id = device_id
|
|
if hasattr(session.asr_service, "session_id"):
|
|
session.asr_service.session_id = session_id
|
|
await establish_asr_connection(session)
|
|
|
|
|
|
async def prepare_llm_service(session, device_id, session_id, selected_role):
|
|
if hasattr(session.llm_service, "device_id"):
|
|
session.llm_service.device_id = device_id
|
|
if hasattr(session.llm_service, "session_id"):
|
|
session.llm_service.session_id = session_id
|
|
if hasattr(session.llm_service, "connect"):
|
|
if not await _connect_with_interrupt_cleanup(session.llm_service, (device_id, session_id)):
|
|
return
|
|
|
|
from services.conversation_history import conversation_history_manager
|
|
from config import settings
|
|
device_history = await conversation_history_manager.get_history(device_id)
|
|
if device_history:
|
|
session.llm_service.history = device_history.history[
|
|
-settings.max_conversation_history :
|
|
]
|
|
|
|
|
|
async def prepare_tts_service(session, device_id, session_id, selected_role):
|
|
session.tts_service.selected_role = selected_role
|
|
if hasattr(session.tts_service, "device_id"):
|
|
session.tts_service.device_id = device_id
|
|
if hasattr(session.tts_service, "session_id"):
|
|
session.tts_service.session_id = session_id
|
|
await establish_tts_connection(session)
|
|
|
|
|
|
async def establish_asr_connection(session: AudioSession):
|
|
|
|
try:
|
|
if session.device_id and session.session_id:
|
|
session_key = (session.device_id, session.session_id)
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
session_logger.info(
|
|
session.device_id, session.session_id, "会话已中断，跳过建立ASR连接"
|
|
)
|
|
return
|
|
session.send_task = await task_manager.create_task(
|
|
send_audio_task(session),
|
|
device_id=session.device_id,
|
|
session_key=(session.device_id, session.session_id) if session.device_id and session.session_id else None,
|
|
task_type="audio_send"
|
|
)
|
|
await session.asr_service.connect()
|
|
if not session.asr_service.connected:
|
|
raise RuntimeError("语音识别 WebSocket 连接未建立")
|
|
if session.device_id and session.session_id:
|
|
if interrupt_handler.is_interrupted((session.device_id, session.session_id)):
|
|
raise RuntimeError("连接期间会话已中断")
|
|
await interrupt_handler.register_cleanup_handler(
|
|
(session.device_id, session.session_id), session.asr_service.close,
|
|
priority=interrupt_handler.PRIORITY["NETWORK_CONNECTION"],
|
|
)
|
|
await interrupt_handler.register_cleanup_handler(
|
|
(session.device_id, session.session_id), session.cancel_send_task,
|
|
priority=interrupt_handler.PRIORITY["TASK_CANCELLATION"],
|
|
)
|
|
session_logger.info(
|
|
"unknown", "unknown", f"语音识别连接已建立，采样率: {session.sample_rate}Hz"
|
|
)
|
|
except asyncio.CancelledError:
|
|
await session.cancel_send_task()
|
|
await session.asr_service.close()
|
|
raise
|
|
except Exception as e:
|
|
session.asr_error = f"{type(e).__name__}: {e}"
|
|
await session.cancel_send_task()
|
|
await session.asr_service.close()
|
|
session_logger.error(
|
|
"unknown", "unknown", f"建立语音识别 WebSocket 连接失败: {e}", exc_info=True
|
|
)
|
|
|
|
|
|
async def establish_tts_connection(session: AudioSession):
|
|
|
|
try:
|
|
if session.device_id and session.session_id:
|
|
session_key = (session.device_id, session.session_id)
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
session_logger.info(
|
|
session.device_id, session.session_id, "会话已中断，跳过建立TTS连接"
|
|
)
|
|
return
|
|
|
|
if not await _connect_with_interrupt_cleanup(
|
|
session.tts_service, (session.device_id, session.session_id)
|
|
):
|
|
return
|
|
|
|
if not session.tts_service.connected:
|
|
session_logger.error(
|
|
session.device_id or "unknown",
|
|
session.session_id or "unknown",
|
|
"语音合成连接未建立",
|
|
)
|
|
return
|
|
|
|
session_logger.info(
|
|
session.device_id or "unknown",
|
|
session.session_id or "unknown",
|
|
"语音合成连接已建立",
|
|
)
|
|
except Exception as e:
|
|
session_logger.error(
|
|
session.device_id or "unknown",
|
|
session.session_id or "unknown",
|
|
f"建立语音合成连接失败: {e}",
|
|
)
|
|
|
|
|
|
async def send_audio_task(session: AudioSession):
|
|
try:
|
|
while True:
|
|
if session.device_id and session.session_id:
|
|
session_key = (session.device_id, session.session_id)
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
session_logger.info(
|
|
session.device_id, session.session_id, "发送音频任务被中断"
|
|
)
|
|
break
|
|
queue = session.send_queue
|
|
audio_data = await queue.get()
|
|
try:
|
|
if audio_data is None: # Drain all preceding audio before finish-task.
|
|
break
|
|
await session.asr_service.send_audio(audio_data)
|
|
finally:
|
|
queue.task_done()
|
|
|
|
except asyncio.CancelledError:
|
|
session.asr_error = session.asr_error or "发送音频任务被取消"
|
|
session_logger.info(
|
|
session.device_id or "unknown",
|
|
session.session_id or "unknown",
|
|
"发送音频数据任务被取消",
|
|
)
|
|
except Exception as e:
|
|
session.asr_error = str(e)
|
|
session_logger.error(
|
|
session.device_id or "unknown",
|
|
session.session_id or "unknown",
|
|
f"发送音频数据时出错: {str(e)}",
|
|
exc_info=True,
|
|
)
|
|
|
|
|
|
async def end_asr_session(session: AudioSession):
|
|
"""Drain device audio, collect the complete transcript, always release ASR."""
|
|
session.transcript = ""
|
|
try:
|
|
session_key = (session.device_id, session.session_id)
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
raise RuntimeError("识别会话已中断")
|
|
if session.send_task and not session.send_task.done():
|
|
await session.send_queue.put(None)
|
|
await asyncio.wait_for(
|
|
asyncio.shield(session.send_task),
|
|
timeout=settings.asr_queue_drain_timeout_seconds,
|
|
)
|
|
if session.asr_error:
|
|
raise RuntimeError(session.asr_error)
|
|
if session.send_queue and not session.send_queue.empty():
|
|
raise RuntimeError("发送任务提前退出，音频未完整发送")
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
raise RuntimeError("识别会话已中断")
|
|
asr_end_start_time = time.perf_counter()
|
|
await session.asr_service.send_end()
|
|
if interrupt_handler.is_interrupted(session_key):
|
|
raise RuntimeError("识别会话已中断")
|
|
session.transcript = await session.asr_service.receive_results()
|
|
session_logger.info(
|
|
session.device_id or "unknown", session.session_id or "unknown",
|
|
f"ASR结束延迟: {time.perf_counter() - asr_end_start_time:.2f} 秒",
|
|
)
|
|
except Exception as e:
|
|
session.asr_error = f"{type(e).__name__}: {e}"
|
|
session.transcript = ""
|
|
session_logger.error(
|
|
session.device_id or "unknown", session.session_id or "unknown",
|
|
f"ASR结束处理出错: {session.asr_error}",
|
|
)
|
|
finally:
|
|
if session.send_task:
|
|
session.send_task.cancel()
|
|
try:
|
|
await asyncio.wait_for(session.send_task, timeout=2.0)
|
|
except (asyncio.CancelledError, asyncio.TimeoutError):
|
|
pass
|
|
finally:
|
|
session.send_task = None
|
|
await session.asr_service.close()
|
|
session_logger.info(
|
|
session.device_id or "unknown", session.session_id or "unknown",
|
|
"语音识别 WebSocket 连接和 ClientSession 已关闭",
|
|
)
|