Is anyone else facing issues with Scribe v2? Getti...
# agents
r
Is anyone else facing issues with Scribe v2? Getting the following errors when trying it locally. It happens when I set the use_realtime flag to true
Copy code
hardcoded_stt = elevenlabs.STT(
            use_realtime=True,
            server_vad={
            "vad_silence_threshold_secs": 0.5,
            "vad_threshold": 0.5,
            "min_speech_duration_ms": 100,
            "min_silence_duration_ms": 300,
        },
        )
Copy code
raise RuntimeError(f"inference of {method} failed: {inf_resp.error}")        
                                     RuntimeError: inference of lk_end_of_utterance_en failed: inference of           
                                     lk_end_of_utterance_en failed: '_EUORunnerEn' object has no attribute            
                                     '_tokenizer'
Copy code
raise APIStatusError(message="ElevenLabs STT connection closed               
                                     unexpectedly")                                                                   
                                     livekit.agents._exceptions.APIStatusError: ElevenLabs STT connection closed      
                                     unexpectedly (status_code=-1, request_id=None, body=None, retryable=True)
l
Are you using the latest version of livekit-agents? This was only very recently added, and I don't see those errors with
1.3.5
r
Working as expected now. Thanks!!
p
I see those errors quite frequently with
1.3.6
deepgram & elevenlabs STT only
is that a known issue? I have upgraded today to
1.3.6
and I am using a FallbackAdapter with both deepgram & elevenlabs
@lively-winter-88668 These are the logs I see.
we are actually an 11labs enterprise customer and this is just the fallback STT - so I dont get why we would get overflow errors
I feel like the WS is not properly closed every time the User stopped speaking / The agent starts speaking.
This is the websocket message that is received before the error shows up:
WSMessage(type=<WSMsgType.CLOSE: 8>, data=1000, extra='')
Have not found a "pattern" does not feel like we hit a timeout or a certain period of silence
l
@plain-librarian-80497 no known issues with this with
1.3.6
"Session terminated: audio data is being sent too frequently. Please reduce the rate at which you send audio chunks"
I'm not sure what is causing this, perhaps try adjusting the sample rate in the STT settings.