fix: 停止帧发送后与原先音频时序问题

回退退出工具打断处理(在大模型思考期间进行打断,然后大模型立马调用退出工具会造成死锁,因模型思考时间和是否调用工具不确定性,状态难管理)
This commit is contained in:
Sakura-RanChen
2026-04-22 15:55:28 +08:00
parent 4acd02e2b0
commit 8c18e48b97
7 changed files with 5 additions and 25 deletions
-6
View File
@@ -148,8 +148,6 @@ class ConnectionHandler:
self.memory = _memory
self.intent = _intent
self.is_exiting = False # 标记是否正在执行退出流程
# 为每个连接单独管理声纹识别
self.voiceprint_provider = None
@@ -350,10 +348,6 @@ class ConnectionHandler:
async def _route_message(self, message):
"""消息路由"""
# 退出状态丢弃所有消息
if self.is_exiting:
return
# 检查是否已经获取到真实的绑定状态
if not self.bind_completed_event.is_set():
# 还没有获取到真实状态,等待直到获取到真实状态或超时
@@ -7,10 +7,6 @@ TAG = __name__
async def handleAbortMessage(conn: "ConnectionHandler"):
if conn.close_after_chat or conn.is_exiting:
conn.logger.bind(tag=TAG).info("退出流程中被打断,直接关闭连接")
return
conn.logger.bind(tag=TAG).info("Abort message received")
# 设置成打断状态,会自动打断llm、tts任务
conn.client_abort = True
@@ -33,10 +33,6 @@ async def handle_user_intent(conn: "ConnectionHandler", text):
if await check_direct_exit(conn, filtered_text):
return True
# 明确再见不被打断
if conn.is_exiting:
return True
# 检查是否是唤醒词
if await checkWakeupWords(conn, filtered_text):
return True
@@ -62,7 +58,6 @@ async def check_direct_exit(conn: "ConnectionHandler", text):
if text == cmd:
conn.logger.bind(tag=TAG).info(f"识别到明确的退出命令: {text}")
await send_stt_message(conn, text)
conn.is_exiting = True
await conn.close()
return True
return False
@@ -15,8 +15,6 @@ TAG = __name__
async def handleAudioMessage(conn: "ConnectionHandler", audio):
if conn.is_exiting:
return
# 当前片段是否有人说话
have_voice = conn.vad.is_vad(conn, audio)
# 如果设备刚刚被唤醒,短暂忽略VAD检测
@@ -84,12 +84,6 @@ class ASRProviderBase(ABC):
async def handle_voice_stop(self, conn: "ConnectionHandler", asr_audio_task: List[bytes]):
"""并行处理ASR和声纹识别"""
try:
# 如果处于退出流程中,直接关闭连接,不处理新消息
if conn.close_after_chat or conn.is_exiting:
logger.bind(tag=TAG).info("退出流程中收到新消息,直接关闭连接")
await conn.close()
return
total_start_time = time.monotonic()
# 准备音频数据
@@ -26,6 +26,7 @@ class ASRProvider(ASRProviderBase):
self.asr_ws = None
self.forward_task = None
self.is_processing = False # 添加处理状态标志
self._is_stopping = False # 添加停止标志,防止竞态条件
# 配置参数
self.appid = str(config.get("appid"))
@@ -148,7 +149,7 @@ class ASRProvider(ASRProviderBase):
return
# 发送当前音频数据
if self.asr_ws and self.is_processing:
if self.asr_ws and self.is_processing and not self._is_stopping:
try:
pcm_frame = self.decoder.decode(audio, 960)
payload = gzip.compress(pcm_frame)
@@ -256,6 +257,7 @@ class ASRProvider(ASRProviderBase):
await self.asr_ws.close()
self.asr_ws = None
self.is_processing = False
self._is_stopping = False
# 重置所有音频相关状态
conn.reset_audio_states()
@@ -264,9 +266,11 @@ class ASRProvider(ASRProviderBase):
asyncio.create_task(self.asr_ws.close())
self.asr_ws = None
self.is_processing = False
self._is_stopping = False
async def _send_stop_request(self):
"""发送最后一个音频帧以通知服务器结束"""
self._is_stopping = True # 先标记为停止状态,阻止后续音频发送
if self.asr_ws:
try:
# 发送结束标记的音频帧(gzip压缩的空数据)
@@ -31,7 +31,6 @@ handle_exit_intent_function_desc = {
"handle_exit_intent", handle_exit_intent_function_desc, ToolType.SYSTEM_CTL
)
def handle_exit_intent(conn: "ConnectionHandler", say_goodbye: str | None = None):
conn.is_exiting = True
# 处理退出意图
try:
if say_goodbye is None: