hey

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

I use beyond presensce avatar, pausing it does not work immediately when used with 11labs tts. is here any open issues?

Thanks — I hear you: pausing your Beyond Presence avatar when using ElevenLabs TTS is not happening immediately, and you want to know if there are any known issues. What I can confirm: • There are no documented or flagged issues in our materials specifically about pausing Beyond Presence avatars when used with ElevenLabs TTS. • Beyond Presence avatars run on a remote synthesis server that publishes synchronized audio and video into the room and support real‑time control (interruptions, RPC events) through LiveKit’s control mechanisms. • ElevenLabs TTS is supported as a TTS provider integration; community/custom voices for LiveKit Inference are not supported and require using the plugin integration instead. Because the docs do not mention this exact pause-delay behavior, let’s narrow the problem so I can help troubleshoot: Please share: 1. Which integration path are you using? ◦ Are you using the Beyond Presence avatar plugin + ElevenLabs plugin, or a custom integration? 1. Where do you observe the delay? ◦ Does the avatar audio continue for a short time after you send a pause command, or does the avatar video continue moving while audio stops? 1. How are you sending the pause command? ◦ From the client (which SDK/language), via LiveKit RPC, or from your server? 1. Example details: ◦ SDK and version, project/room ID (or any anonymized IDs), timestamps when this happens, and any console/log errors. 1. Repro steps and frequency: ◦ Can you consistently reproduce it? If so, short step-by-step reproduction helps a lot. Immediate troubleshooting suggestions you can try now: • Confirm you’re using RPC/events that the avatar expects for interruption or pause (the avatar expects real‑time control events rather than only manipulating audio tracks). • If the avatar’s audio is published from a remote server, check whether the pause command is reaching that synthesis server (server logs) and whether ElevenLabs TTS playback has internal buffering or latency. • Test with a different TTS provider or with a simple/local TTS to see whether the delay is specific to ElevenLabs audio output. • Capture a short log or network trace around the pause command to see timing between command send, command receipt, and audio/video stop. If you send me the answers to the questions above and any short logs or repro steps, I’ll walk through targeted next steps — for example, verifying the RPC event payloads, checking for buffering behavior on the TTS side, or suggesting a change to how we interrupt playback to make pausing immediate. Sources: ElevenLabs TTS | LiveKit Documentation | Beyond Presence virtual avatar integration guide | LiveKit Documentation | examples/avatar_agents | livekit/livekit_composite | DeepWiki | Bringing AI avatars to voice agents

How to check agent observability logs ?

Thanks — I understand you want to check agent observability logs to help troubleshoot your avatar + TTS behavior. I’ll walk you through how to access them and what to check. Quick checklist (prerequisites) • Agent Observability is enabled for your project under Data and privacy. • Your agent uses a supported SDK: Python >= 1.3.0 or Node.js >= 1.0.18. • The session wasn’t created with recording disabled (e.g.,

record=False

), and the agent was able to shut down cleanly so it could upload logs. How to view Agent observability logs in the dashboard 1. Go to your project’s sessions dashboard: ◦ Open the Sessions page for your project. 1. Select the session you want to inspect. 2. Open the Agent insights tab, then the Logs sub-tab. ◦ Logs are shown chronologically for that session. 1. Use “Download data” to save logs as JSON or “Share” to provide them to others. Important details about the logs • Logs are session-scoped and retained for 30 days. • Logs are not streamed live to the dashboard: data appears after the session ends and may take up to ~5 minutes to process. • Observability data is uploaded at session end. If an agent crashes during shutdown, the upload may fail. If you don’t see logs (quick troubleshooting) • Confirm Observability is enabled in project settings (Data and privacy). • Check agent runtime/debug logs for a message like "uploading session report to LiveKit Cloud" to verify upload succeeded. • Ensure the agent used a supported SDK version and the session allowed recording. • If the agent crashed before upload, locate local agent logs (server logs) to capture immediate errors. For longer-term or live monitoring • Consider forwarding agent logs to Datadog, CloudWatch, Sentry, or New Relic (supported for agents hosted on LiveKit Cloud). If you’d like, share: • Which project and session (anonymized IDs are fine), • SDK and version you’re using, • Whether you see any upload/agent errors in your server logs. With those details I can help confirm whether observability was uploaded correctly and point to specific log entries that will help diagnose the pause/delay issue. Sources: Streamline troubleshooting with Agent Observability | Why is my Agent insights data missing? | LiveKit | Agent Builder | LiveKit Documentation | Where to find your Agent Logs | LiveKit

what is this? resume_false_interruption is enabled but audio output does not support pause, it will be ignored

Thanks — I can explain what that warning means and how to fix it. What the message means • You have

resume_false_interruption=True

enabled on your AgentSession, but the audio output currently in use does not implement pause/resume capability. • Because the audio output can’t actually pause, the resume-on-false-interruption behavior will be ignored. The agent won’t be able to temporarily pause playback when it detects possible user speech and then resume from the same position if the speech turns out to be a false interruption. Why this happens (common causes) • Avatar audio outputs (like Beyond Presence avatar integrations) often publish synthesized audio/video from a remote server and don’t support `pause()`/`resume()` — so

can_pause

is effectively false. • A custom audio output you’re using may not implement pause/resume. • Some TTS/playback chains buffer audio in a way that prevents precise pausing/resuming. What you can do (options) • Disable the feature if your audio path can’t pause: ◦ Set

resume_false_interruption=False

on the AgentSession. This will make false interruptions behave like normal interruptions (the agent will not automatically resume). • Use an audio output that supports pause/resume: ◦ Replace the avatar audio output or TTS playback path with one that implements pause() and resume() and reports

can_pause=True

. • Workaround testing: ◦ Test with a simple/local TTS output that supports pause/resume to confirm behavior. ◦ If you must keep the avatar, treat false interruptions as regular interruptions (disable

resume_false_interruption

) to avoid unpredictable behavior. Impact of disabling resume_false_interruption • The agent will not resume playback after a false interruption; it will wait for new user input instead. This is the recommended approach when the audio output cannot pause. If you want, tell me: • Which audio output you’re using (avatar plugin + ElevenLabs TTS, a custom audio output, etc.) • SDK and AgentSession settings (snippet with

resume_false_interruption

and

false_interruption_timeout

) • Any relevant logs showing the warning With that I can recommend the exact change (code snippet) or help confirm whether your current audio path can support pause/resume. Sources: Turns overview | LiveKit Documentation | How to Handle Speech Interruptions More Naturally