Google STT times out at 5 minutes and crashes agent

[nickderobertis]

Google STT seems to consistently fail at 5 minutes into the call with the following error:

2024-12-18 15:59:58,786 - ERROR livekit.agents.pipeline - Error in _recognize_task
Traceback (most recent call last):
  File "/Users/nick/repos/hellopatient/lk_agent/.venv/lib/python3.12/site-packages/livekit/agents/utils/log.py", line 16, in async_fn_logs
    return await fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/nick/repos/hellopatient/lk_agent/.venv/lib/python3.12/site-packages/livekit/agents/pipeline/human_input.py", line 150, in _recognize_task
    await asyncio.gather(*tasks)
  File "/Users/nick/repos/hellopatient/lk_agent/.venv/lib/python3.12/site-packages/livekit/agents/pipeline/human_input.py", line 120, in _audio_stream_co
    stt_stream.push_frame(ev.frame)
  File "/Users/nick/repos/hellopatient/lk_agent/.venv/lib/python3.12/site-packages/livekit/agents/stt/stt.py", line 265, in push_frame
    self._check_not_closed()
  File "/Users/nick/repos/hellopatient/lk_agent/.venv/lib/python3.12/site-packages/livekit/agents/stt/stt.py", line 327, in _check_not_closed
    raise RuntimeError(f"{cls.__module__}.{cls.__name__} is closed")
RuntimeError: livekit.plugins.google.stt.SpeechStream is closed {"pid": 54793, "job_id": "AJ_zKEGnh4r4ttx"}

Unfortunately it's an unhandled error and so even though we were using it with in the FallbackAdapter our agent stops responding completely at 5 minutes into the call.

I have worked with the Google API before and I know it times out the streaming connection at 5 minutes so I imagine that is the issue. There's an official tutorial on endless streaming that shows how to switch to a new request and stitch them together in real-time.

[potate4]

same issue