how to capture end of utterance delay in agent LiveKit Community #ask-ai

Join Slack

how to capture end-of-utterance delay in agent?

# ask-ai

billions-lawyer-8588

08/12/2025, 3:50 PM

how to capture end-of-utterance delay in agent?

tall-airline-16756

08/12/2025, 3:50 PM

Thanks for reaching out, Harsh. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

tall-airline-16756

08/12/2025, 3:51 PM

Hey there! If you want to capture the end-of-utterance (EOU) delay in a LiveKit agent, you can use the

PipelineEOUMetrics

interface. It has this handy

endOfUtteranceDelay

property that shows the time between when speech ends (detected by VAD) and when the agent decides to wrap up the user's turn. Here’s a simple way to get to it: 1. Metric Collection: The

endOfUtteranceDelay

pops up as part of the

EOUMetrics

right before the agent plays speech back to the user. Just listen for the

METRICS_COLLECTED

event to grab this info. 2. Example:

Copy code

def eou_metrics_wrapper(metrics: EOUMetrics):
    eou_delay = metrics.end_of_utterance_delay  # Here's the delay
    transcription_delay = metrics.transcription_delay
    # Do whatever you need with these metrics

agent.eou_metrics_event.subscribe(eou_metrics_wrapper)

This way, you can easily capture and use the EOU delay for any analysis or processing you have in mind. If you want to dive deeper, check out the LiveKit documentation on metrics and the Python examples repository. They’ve got some great info to help you along! Sources: Capturing metrics | LiveKit Docs | complex-agents/turn-taking/agent.py

billions-lawyer-8588

08/12/2025, 5:08 PM

Does session have any decorator for this, like we have to capture stt, tts and llm metrics?

tall-airline-16756

08/12/2025, 5:08 PM

Hey there! Just a quick heads-up: LiveKit doesn’t have a dedicated session decorator for capturing end-of-utterance (EOU) delay metrics like it does for STT, TTS, and LLM metrics. But no worries—you can still grab those EOU metrics, including

end_of_utterance_delay

and

transcription_delay

, by subscribing to the

eou_metrics_collected

event on the STT component. Here’s a little example to get you started:

Copy code

def eou_metrics_wrapper(metrics: EOUMetrics):
    asyncio.create_task(self.on_eou_metrics_collected(metrics))

stt.on("eou_metrics_collected", eou_metrics_wrapper)

These metrics come from the

EOUMetrics

interface and you can log or analyze them however you like. If you want more details or another example, just give me a shout—I’m here to help! Sources: PipelineEOUMetrics | LiveKit Agents | metrics/metrics_stt.py | livekit-agents/livekit/agents/voice/agent_session.py

2 Views

Open in Slack

Previous Next