Thanks — I’m sorry you’re stuck on “Waiting for agent audio track.” I’ll help troubleshoot. A few likely causes and quick checks: What I’m responding to: your Speechmatics + diarisation works locally in console but when deployed the playground shows “Waiting for agent audio track.” Quick checks (please confirm results): • Is TTS/audio output enabled in your agent? Verify

session.output.set_audio_enabled(True)

or that you call

session.say()

/

session.generate_reply()

in the deployed code. • Did you deploy your Speechmatics API key as a secret? Run

lk agent secrets list

and confirm

SPEECHMATICS_API_KEY

is present. • Are you running text-only mode in deployment? Text-only or disabled audio will not publish an audio track. • Any errors in logs when the agent connects? Please share the output of

lk agent logs

around the time the room connects. If those are OK, try these targeted debugging steps: 1. Add logging to confirm audio publishing:

@session.on("agent_state_changed")
   def on_agent_state_change(ev):
       print(f"Agent state: {ev.new_state}")

1. Force a simple spoken reply to test audio publishing:

await session.say("Hello", allow_interruptions=False)

1. Check room participants / published tracks (look for agent

AudioTrack

) or observe

track_published

events. If you share: • a short snippet of your deployment entrypoint (where you configure output/TTS and call

session.say()

), • the

lk agent logs

lines when the agent connects, I can point to the exact setting to change. Sources: Google AI and LiveKit | LiveKit Docs | Text and transcriptions | LiveKit docs | Speechmatics STT plugin guide | LiveKit docs | Agents (Python) | livekit/livekit_composite | DeepWiki | How to detect when an agent has finished speaking