Run VAD on all incoming audio frames

bridge/app.py

@@ -409,6 +409,8 @@ async def handle_text_message(device_id: str, session: DeviceSession, raw: str)
 
 async def handle_binary_message(device_id: str, session: DeviceSession, data: bytes) -> None:
     session.last_rx_ts = time.time()
+    # VAD should work on continuous stream even if explicit start/stop is missing.
+    process_vad_frame(device_id, session, data)
     if session.ptt_active:
         session.rx_bytes_total += len(data)
         if SAVE_SESSIONS:
@@ -419,7 +421,6 @@ async def handle_binary_message(device_id: str, session: DeviceSession, data: by
                 # Keep newest data within cap to avoid unbounded memory growth.
                 drop = len(session.pcm_bytes) - MAX_SESSION_BYTES
                 del session.pcm_bytes[:drop]
-        process_vad_frame(device_id, session, data)
     if ECHO_ENABLED:
         await session.ws.send(data)