Merge pull request #1640 from xinnan-tech/py_tts_listen

fix: 反复打断任务没有被清除，vad四帧语音识别

main/xiaozhi-server/config/logger.py

@@ -5,7 +5,7 @@ from config.config_loader import load_config
 from config.settings import check_config_file
 from datetime import datetime
 
-SERVER_VERSION = "0.5.7"
+SERVER_VERSION = "0.5.8"
 _logger_initialized = False
 
 

main/xiaozhi-server/core/connection.py

@@ -115,6 +115,7 @@ class ConnectionHandler:
         self.client_have_voice_last_time = 0.0
         self.client_no_voice_last_time = 0.0
         self.client_voice_stop = False
+        self.client_voice_frame_count = 0
 
         # asr相关变量
         # 因为实际部署时可能会用到公共的本地ASR，不能把变量暴露给公共ASR
@@ -627,8 +628,8 @@ class ConnectionHandler:
                 )
                 memory_str = future.result()
 
-            uuid_str = str(uuid.uuid4()).replace("-", "")
+            self.sentence_id = str(uuid.uuid4().hex)
-            self.sentence_id = uuid_str
+
 
             if self.intent_type == "function_call" and functions is not None:
                 # 使用支持functions的streaming接口

main/xiaozhi-server/core/providers/tts/huoshan_double_stream.py

@@ -12,6 +12,8 @@ from core.utils.util import check_model_key
 from core.providers.tts.base import TTSProviderBase
 from core.handle.abortHandle import handleAbortMessage
 from core.providers.tts.dto.dto import SentenceType, ContentType, InterfaceType
+from asyncio import Task
+
 
 TAG = __name__
 logger = setup_logging()
@@ -141,6 +143,7 @@ class TTSProvider(TTSProviderBase):
         super().__init__(config, delete_audio_file)
         self.ws = None
         self.interface_type = InterfaceType.DUAL_STREAM
+        self._monitor_task = None  # 监听任务引用
         self.appId = config.get("appid")
         self.access_token = config.get("access_token")
         self.cluster = config.get("cluster")
@@ -270,8 +273,7 @@ class TTSProvider(TTSProviderBase):
         try:
             # 建立新连接
             if self.ws is None:
-                await handleAbortMessage(self.conn)
+                logger.bind(tag=TAG).warning(f"WebSocket连接不存在，终止发送文本")
-                logger.bind(tag=TAG).error(f"WebSocket连接不存在，终止发送文本")
                 return
 
             #  过滤Markdown
@@ -293,6 +295,25 @@ class TTSProvider(TTSProviderBase):
     async def start_session(self, session_id):
         logger.bind(tag=TAG).info(f"开始会话～～{session_id}")
         try:
+            task = self._monitor_task
+            if (
+                task is not None
+                and isinstance(task, Task)
+                and not task.done()
+            ):
+                logger.bind(tag=TAG).info("等待上一个监听任务结束...")
+                if self.ws is not None:
+                    logger.bind(tag=TAG).info("强制关闭上一个WebSocket连接以唤醒监听任务...")
+                    try:
+                        await self.ws.close()
+                    except Exception as e:
+                        logger.bind(tag=TAG).warning(f"关闭上一个ws异常: {e}")
+                    self.ws = None
+                try:
+                    await asyncio.wait_for(task, timeout=8)
+                except Exception as e:
+                    logger.bind(tag=TAG).warning(f"等待监听任务异常: {e}")
+                self._monitor_task = None
             # 建立新连接
             await self._ensure_connection()
 
@@ -463,6 +484,8 @@ class TTSProvider(TTSProviderBase):
                 except:
                     pass
                 self.ws = None
+            # 监听任务退出时清理引用
+            self._monitor_task = None
 
     async def send_event(
         self,

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

@@ -35,6 +35,10 @@ class VADProvider(VADProviderBase):
             pcm_frame = self.decoder.decode(opus_packet, 960)
             conn.client_audio_buffer.extend(pcm_frame)  # 将新数据加入缓冲区
 
+            # 初始化帧计数器
+            if not hasattr(conn, "client_voice_frame_count"):
+                conn.client_voice_frame_count = 0
+
             # 处理缓冲区中的完整帧（每次处理512采样点）
             client_have_voice = False
             while len(conn.client_audio_buffer) >= 512 * 2:
@@ -50,7 +54,15 @@ class VADProvider(VADProviderBase):
                 # 检测语音活动
                 with torch.no_grad():
                     speech_prob = self.model(audio_tensor, 16000).item()
-                client_have_voice = speech_prob >= self.vad_threshold
+                is_voice = speech_prob >= self.vad_threshold
+
+                if is_voice:
+                    conn.client_voice_frame_count += 1
+                else:
+                    conn.client_voice_frame_count = 0
+
+                # 只有连续4帧检测到语音才认为有语音
+                client_have_voice = conn.client_voice_frame_count >= 4
 
                 # 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: