Elevenlabs TTS issue in the latest pipecat version

[Shoshin23]

Hi,

We're using Pipecat with Fly.io. With the latest version (0.0.42), the ElevenLabs TTS seems to be throwing an error constantly, which is stopping the machines from auto-destroying.

I'm putting the error here along with this ticket:

2024-10-08T15:31:45.684 app[3d8d3e2f769568] ams [info] 2024-10-08 15:31:45.683 | ERROR | pipecat.services.elevenlabs:_keepalive_task_handler:372 - ElevenLabsTTSService#0 exception: no close frame received or sent

2024-10-08T15:31:47.080 app[148e2444f23768] ams [info] 2024-10-08 15:31:47.079 | ERROR | pipecat.services.elevenlabs:_keepalive_task_handler:372 - ElevenLabsTTSService#0 exception: no close frame received or sent

2024-10-08T15:31:47.550 app[56830971f5d4d8] ams [info] 2024-10-08 15:31:47.549 | ERROR | pipecat.services.elevenlabs:_keepalive_task_handler:372 - ElevenLabsTTSService#0 exception: received 1012 (service restart); then sent 1012 (service restart)

Could you please help us troubleshoot it?

This is also a sample of my implementation:

    async with aiohttp.ClientSession() as session:
        transport = DailyTransport(
            room_url,
            token,
            "Chatbot",
            DailyParams(
                api_url=daily_api_url,
                api_key=daily_api_key,
                audio_in_enabled=True,
                audio_out_enabled=True,
                camera_out_enabled=True,
                camera_out_width=1024,
                camera_out_height=576,
                vad_enabled=True,
                vad_analyzer=SileroVADAnalyzer(),
                transcription_enabled=True,
            )
        )

        tts = ElevenLabsTTSServiceWithCleanup(
            aiohttp_session=session,
            api_key=os.getenv("ELEVENLABS_API_KEY", ""),
            voice_id=voice,)

        llm = OpenAILLMService(
            api_key=os.getenv("OPENAI_API_KEY"),
            model="gpt-4o")

        sa = SentenceAggregator()
        va = VisionImageFrameAggregator()
        tma_in = LLMUserResponseAggregator(messages)
        tma_out = LLMAssistantResponseAggregator(messages)
        llm_debugger = LLMDebugger()
        transcription_debugger = TranscriptionDebugger()
        ir = UserImageRequester()
        tf = TextFilterProcessor(user_request_answer)
        imgf = ImageFilterProcessor()

        # ura = LLMUserResponseAggregator(messages)
        
        ura = LastQuestionCaptureAggregator(messages, ir)

        pipeline = Pipeline(
            [
                transport.input(),
                ura,
                llm,
                ParallelPipeline([sa, ir, va, llm], [tf, imgf]),
                tts,
                transport.output(),
            ]
        ) 

task = PipelineTask(pipeline, PipelineParams(allow_interruptions=True))

We're not doing anything specific with Elevenlabs apart from just calling it or passing it to the pipeline task. Happy to share more information, if needed.

[danthegoodman1]

getting this as well

[markbackman]

@aconchillo made changes to improve the reconnection logic of the ElevenLabs websocket: #796. That should resolve this issue. If you still see this problem, please open a new issue.