update:长按说话不走VAD直接触发ASR识别

main/xiaozhi-server/core/handle/textHandler/listenMessageHandler.py

@@ -30,7 +30,13 @@ class ListenTextMessageHandler(TextMessageHandler):
             conn.client_have_voice = True
             conn.client_voice_stop = True
             if len(conn.asr_audio) > 0:
-                await handleAudioMessage(conn, b"")
+                # 手动模式下直接触发ASR识别，不需要再调用handleAudioMessage
+                asr_audio_task = conn.asr_audio.copy()
+                conn.asr_audio.clear()
+                conn.reset_vad_states()
+                
+                if len(asr_audio_task) > 0:
+                    await conn.asr.handle_voice_stop(conn, asr_audio_task)
         elif msg_json["state"] == "detect":
             conn.client_have_voice = False
             conn.asr_audio.clear()

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

@@ -56,21 +56,28 @@ class ASRProviderBase(ABC):
     async def receive_audio(self, conn, audio, audio_have_voice):
         if conn.client_listen_mode == "auto" or conn.client_listen_mode == "realtime":
             have_voice = audio_have_voice
+            
+            conn.asr_audio.append(audio)
+            if not have_voice and not conn.client_have_voice:
+                conn.asr_audio = conn.asr_audio[-10:]
+                return
         else:
-            have_voice = conn.client_have_voice
-        
-        conn.asr_audio.append(audio)
-        if not have_voice and not conn.client_have_voice:
-            conn.asr_audio = conn.asr_audio[-10:]
-            return
+            # 手动模式：总是缓存音频，忽略VAD检测结果
+            conn.asr_audio.append(audio)
 
         if conn.client_voice_stop:
             asr_audio_task = conn.asr_audio.copy()
             conn.asr_audio.clear()
             conn.reset_vad_states()
 
-            if len(asr_audio_task) > 15:
-                await self.handle_voice_stop(conn, asr_audio_task)
+            # 手动模式下允许短语音识别，自动模式保持原有限制
+            if conn.client_listen_mode == "auto" or conn.client_listen_mode == "realtime":
+                if len(asr_audio_task) > 15:
+                    await self.handle_voice_stop(conn, asr_audio_task)
+            else:
+                # 手动模式：只要有音频就进行识别
+                if len(asr_audio_task) > 0:
+                    await self.handle_voice_stop(conn, asr_audio_task)
 
     # 处理语音停止
     async def handle_voice_stop(self, conn, asr_audio_task: List[bytes]):

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

@@ -45,6 +45,10 @@ class VADProvider(VADProviderBase):
                 pass
 
     def is_vad(self, conn, opus_packet):
+        # 手动模式：直接返回True，不进行实时VAD检测，所有音频都缓存
+        if conn.client_listen_mode not in ["auto", "realtime"]:
+            return True
+            
         try:
             pcm_frame = self.decoder.decode(opus_packet, 960)
             conn.client_audio_buffer.extend(pcm_frame)  # 将新数据加入缓冲区