The bot disconnects the call before speaking the final sentence.

[devedgecio]

I am using the following pipecat version.

"pipecat-ai==0.0.58"

Issue:

My end_call function executes immediately before the bot says 'Thanks for calling! Goodbye'. I have tried two different approaches to end the call but got the same issue! Moreover, there is a possibility that the production application is handling more than one call at a time. If the issue can be resolved using the second approach, that would be ideal.

Using Pipecat

I used the following code:

async def end_call_handler(function_name, toll_call_id, args, llm, context, result_callback):
    # print(f"Hanging up call for client ID: {args['callSid']}")
    await llm.push_frame(EndTaskFrame(), FrameDirection.UPSTREAM)


async def callback_end_call(function_name, llm, context):
    await llm.push_frame(TTSSpeakFrame("Thanks for Calling, Goodbye!"))
    logger.debug(f"Ending call with function_name: {function_name}")

llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")
llm.register_function("end_call", end_call_handler, start_callback = callback_end_call)

Issues with pipecat:

When bot calls this function, it basically disconnects the Deepgram, cancel the pipeline tasks and throws an error.

logs of error:

2025-03-19 16:15:40,389 - INFO - HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
2025-03-19 16:15:40.589 | DEBUG    | bot:callback_end_call:128 - Ending call with function_name: end_call
2025-03-19 16:15:40.590 | INFO     | pipecat.processors.aggregators.openai_llm_context:call_function:192 - Calling function end_call with arguments {'callSid': 'CA7b9f35d34e273d44feb62484b7bee635'}
2025-03-19 16:15:40.590 | DEBUG    | pipecat.services.openai:process_frame:604 - FunctionCallInProgressFrame: FunctionCallInProgressFrame#0
2025-03-19 16:15:40.590 | DEBUG    | pipecat.services.deepgram:run_tts:73 - Generating TTS: [Thanks for Calling Legal Services, Goodbye!]
2025-03-19 16:15:40.591 | DEBUG    | pipecat.services.deepgram:_disconnect:229 - Disconnecting from Deepgram
tasks cancelled error: 
2025-03-19 16:15:41,093 - ERROR - tasks cancelled error: 
tasks cancelled error: 
2025-03-19 16:15:41,093 - ERROR - tasks cancelled error: 
2025-03-19 16:15:41.946 | DEBUG    | pipecat.pipeline.task:cancel:218 - Canceling pipeline task PipelineTask#0
INFO:     connection closed
2025-03-19 16:15:41.947 | DEBUG    | pipecat.services.deepgram:_disconnect:229 - Disconnecting from Deepgram
tasks cancelled error: 
2025-03-19 16:15:41,947 - ERROR - tasks cancelled error: 
tasks cancelled error: 
2025-03-19 16:15:41,948 - ERROR - tasks cancelled error: 
2025-03-19 16:15:41.988 | DEBUG    | pipecat.pipeline.runner:_gc_collect:75 - Garbage collector: collected 108 objects.
2025-03-19 16:15:41.988 | DEBUG    | pipecat.pipeline.runner:_gc_collect:76 - Garbage collector: uncollectable objects []
2025-03-19 16:15:41.988 | DEBUG    | pipecat.pipeline.runner:run:50 - Runner PipelineRunner#0 finished running PipelineTask#0

Using Twilio with callSid:

I used the following code:

async def end_call_handler(function_name, tool_call_id, args, llm, context, result_callback):
    call_sid = args.get("callSid")
    url = f"https://api.twilio.com/2010-04-01/Accounts/{account_sid}/Calls/{call_sid}.json"
    data = {"Status": "completed"}

    response = requests.post(
        url,
        data=data,
        auth=HTTPBasicAuth(account_sid, auth_token),
    )

    # Check the response
    if response.status_code == 200:
        print(f"Call ended successfully: {call_sid}. {response.text}")
        result = "call has been ended successfully!"
    else:
        print(f"Error ending call: {response.status_code} - {response.text}")
        result = "Issue in Call ending"
    # await llm.push_frame(EndTaskFrame(), FrameDirection.UPSTREAM)
    properties = FunctionCallResultProperties(run_llm=False)
    await result_callback(result, properties=properties)

async def callback_end_call(function_name, llm, context):
    await llm.push_frame(TTSSpeakFrame("Thanks for Calling, Goodbye!"))
    logger.debug(f"Ending call with function_name: {function_name}")

llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")
llm.register_function("end_call", end_call_handler, start_callback = callback_end_call)

Issue:

I got the same issue. Bot doesn't speak the last line "Thanks for calling! Goodbye!"

logs:

2025-03-19 16:24:44,542 - INFO - HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
2025-03-19 16:24:44.765 | DEBUG    | bot:callback_end_call:129 - Ending call with function_name: end_call
2025-03-19 16:24:44.765 | INFO     | pipecat.processors.aggregators.openai_llm_context:call_function:192 - Calling function end_call with arguments {'callSid': 'CA07d7e7838bbdd8395a682b58c226060c'}
2025-03-19 16:24:44.765 | DEBUG    | pipecat.services.openai:process_frame:604 - FunctionCallInProgressFrame: FunctionCallInProgressFrame#0
Call ended successfully: CA07d7e7838bbdd8395a682b58c226060c.
2025-03-19 16:24:45.752 | DEBUG    | pipecat.services.deepgram:run_tts:73 - Generating TTS: [Thanks for Calling Legal Services, Goodbye!]
2025-03-19 16:24:46.985 | DEBUG    | pipecat.pipeline.task:cancel:218 - Canceling pipeline task PipelineTask#0
INFO:     connection closed
2025-03-19 16:24:46.986 | DEBUG    | pipecat.services.deepgram:_disconnect:229 - Disconnecting from Deepgram
tasks cancelled error: 
2025-03-19 16:24:47,488 - ERROR - tasks cancelled error: 
tasks cancelled error: 
2025-03-19 16:24:47,489 - ERROR - tasks cancelled error: 
2025-03-19 16:24:47.528 | DEBUG    | pipecat.pipeline.runner:_gc_collect:75 - Garbage collector: collected 97 objects.
2025-03-19 16:24:47.528 | DEBUG    | pipecat.pipeline.runner:_gc_collect:76 - Garbage collector: uncollectable objects []
2025-03-19 16:24:47.528 | DEBUG    | pipecat.pipeline.runner:run:50 - Runner PipelineRunner#0 finished running PipelineTask#0

[sphatate]

Yes, this is a issue we are also facing. As the end message and function call happens simultaneously. The bot / call is disconnect and the end msg is never spoken

[sphatate]

Can we do something like is TTS Frame is still going we wait and once everything is finished then the call is ended

[devedgecio]

@markbackman, I opened it as a separate issue! Is there any update on this?

[aconchillo]

Hi! I believe we fixed this issue in 0.0.62. Any chance you can try again?