refactor: vad独立时间戳，相关残余状态处理

main/xiaozhi-server/core/connection.py

@@ -159,6 +159,7 @@ class ConnectionHandler:
         self.client_voice_window = deque(maxlen=5)
         self.first_activity_time = 0.0  # 记录首次活动的时间（毫秒）
         self.last_activity_time = 0.0  # 统一的活动时间戳（毫秒）
+        self.vad_last_voice_time = 0.0  # 记录用户最后一次说话的时间（毫秒）
         self.client_voice_stop = False
         self.last_is_voice = False
 
@@ -1288,10 +1289,7 @@ class ConnectionHandler:
             # 标记任务完成
             self.report_queue.task_done()
 
-    def clearSpeakStatus(self, sentence_id=None):
-        # 如果sentence_id不匹配，说明是旧轮次的回调，不需要执行
-        if sentence_id is not None and sentence_id != self.sentence_id:
-            return
+    def clearSpeakStatus(self):
         self.client_is_speaking = False
         self.logger.bind(tag=TAG).debug(f"清除服务端讲话状态")
 
@@ -1433,6 +1431,7 @@ class ConnectionHandler:
         self.client_voice_stop = False
         self.client_voice_window.clear()
         self.last_is_voice = False
+        self.vad_last_voice_time = 0.0
 
         # Clear ASR buffers
         self.asr_audio.clear()

main/xiaozhi-server/core/handle/receiveAudioHandle.py

@@ -77,6 +77,7 @@ async def startToChat(conn: "ConnectionHandler", text):
         ):
             await max_out_size(conn)
             return
+
     # manual 模式下不打断正在播放的内容
     if conn.client_is_speaking and conn.client_listen_mode != "manual":
         await handleAbortMessage(conn)

main/xiaozhi-server/core/handle/sendAudioHandle.py

@@ -284,8 +284,10 @@ async def send_tts_message(conn: "ConnectionHandler", state, text=None):
         # 停止音频发送循环（仅在流控器已初始化时调用）
         if hasattr(conn, "audio_rate_controller") and conn.audio_rate_controller:
             conn.audio_rate_controller.stop_sending()
-        # 清除服务端讲话状态，传入 sentence_id 用于判断是否是当前轮次
-        conn.clearSpeakStatus(current_sentence_id)
+        # 检查是否是当前轮次
+        if current_sentence_id != conn.sentence_id:
+            return
+        conn.clearSpeakStatus()
 
     # 发送消息到客户端
     await conn.websocket.send(json.dumps(message))

main/xiaozhi-server/core/providers/asr/base.py

@@ -84,6 +84,12 @@ 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

main/xiaozhi-server/core/providers/vad/silero.py

@@ -107,12 +107,12 @@ class VADProvider(VADProviderBase):
 
                 # 如果之前有声音，但本次没有声音，且与上次有声音的时间差已经超过了静默阈值，则认为已经说完一句话
                 if conn.client_have_voice and not client_have_voice:
-                    stop_duration = time.time() * 1000 - conn.last_activity_time
+                    stop_duration = time.time() * 1000 - conn.vad_last_voice_time
                     if stop_duration >= self.silence_threshold_ms:
                         conn.client_voice_stop = True
                 if client_have_voice:
                     conn.client_have_voice = True
-                    conn.last_activity_time = time.time() * 1000
+                    conn.vad_last_voice_time = time.time() * 1000
 
             return client_have_voice
         except opuslib_next.OpusError as e: