diff --git a/main/xiaozhi-server/core/connection.py b/main/xiaozhi-server/core/connection.py
index 4337a488..92043e9b 100644
--- a/main/xiaozhi-server/core/connection.py
+++ b/main/xiaozhi-server/core/connection.py
@@ -115,9 +115,9 @@ class ConnectionHandler:
         # vad相关变量
         self.client_audio_buffer = bytearray()
         self.client_have_voice = False
+        self.client_voice_window = deque(maxlen=5)
         self.last_activity_time = 0.0  # 统一的活动时间戳（毫秒）
         self.client_voice_stop = False
-        self.client_voice_window = deque(maxlen=5)
         self.last_is_voice = False
 
         # asr相关变量
diff --git a/main/xiaozhi-server/core/handle/receiveAudioHandle.py b/main/xiaozhi-server/core/handle/receiveAudioHandle.py
index e7b6a138..4eaf9ab1 100644
--- a/main/xiaozhi-server/core/handle/receiveAudioHandle.py
+++ b/main/xiaozhi-server/core/handle/receiveAudioHandle.py
@@ -33,7 +33,7 @@ async def handleAudioMessage(conn, audio):
 
 async def resume_vad_detection(conn):
     # 等待2秒后恢复VAD检测
-    await asyncio.sleep(1)
+    await asyncio.sleep(2)
     conn.just_woken_up = False
 
 
diff --git a/main/xiaozhi-server/core/providers/asr/base.py b/main/xiaozhi-server/core/providers/asr/base.py
index 250a25f2..2562f589 100644
--- a/main/xiaozhi-server/core/providers/asr/base.py
+++ b/main/xiaozhi-server/core/providers/asr/base.py
@@ -68,7 +68,7 @@ class ASRProviderBase(ABC):
             conn.asr_audio.clear()
             conn.reset_vad_states()
 
-            if len(asr_audio_task) > 15:
+            if len(asr_audio_task) > 15 or conn.client_listen_mode == "manual":
                 await self.handle_voice_stop(conn, asr_audio_task)
 
     # 处理语音停止
diff --git a/main/xiaozhi-server/core/providers/vad/silero.py b/main/xiaozhi-server/core/providers/vad/silero.py
index b516d8fb..2263fcb7 100644
--- a/main/xiaozhi-server/core/providers/vad/silero.py
+++ b/main/xiaozhi-server/core/providers/vad/silero.py
@@ -70,7 +70,9 @@ class VADProvider(VADProviderBase):
 
                 # 更新滑动窗口
                 conn.client_voice_window.append(is_voice)
-                client_have_voice = (conn.client_voice_window.count(True) >= self.frame_window_threshold)
+                client_have_voice = (
+                    conn.client_voice_window.count(True) >= self.frame_window_threshold
+                )
 
                 # 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: