allow pushing frames to VAD when agent speech is uninterruptible (#4418)

Author: chenghao-mou

livekit-agents/livekit/agents/voice/agent_activity.py

@@ -735,19 +735,20 @@ def push_audio(self, frame: rtc.AudioFrame) -> None:
         if not self._started:
             return
 
-        if (
+        should_discard = bool(
             self._current_speech
             and not self._current_speech.allow_interruptions
             and self._session.options.discard_audio_if_uninterruptible
-        ):
-            # discard the audio if the current speech is not interruptable
-            return
+        )
 
-        if self._rt_session is not None:
-            self._rt_session.push_audio(frame)
+        if not should_discard:
+            if self._rt_session is not None:
+                self._rt_session.push_audio(frame)
 
+        # Always forward to _audio_recognition for VAD, even when discarding STT/LLM
+        # VAD needs frames to detect speech end and update user state correctly
         if self._audio_recognition is not None:
-            self._audio_recognition.push_audio(frame)
+            self._audio_recognition.push_audio(frame, skip_stt=should_discard)
 
     def push_video(self, frame: rtc.VideoFrame) -> None:
         if not self._started:

livekit-agents/livekit/agents/voice/audio_recognition.py

@@ -180,9 +180,9 @@ def stop(self) -> None:
         self.update_stt(None)
         self.update_vad(None)
 
-    def push_audio(self, frame: rtc.AudioFrame) -> None:
+    def push_audio(self, frame: rtc.AudioFrame, *, skip_stt: bool = False) -> None:
         self._sample_rate = frame.sample_rate
-        if self._stt_ch is not None:
+        if not skip_stt and self._stt_ch is not None:
             self._stt_ch.send_nowait(frame)
 
         if self._vad_ch is not None: