when is on_user_turn_completed function invoked ?

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

to detect speech earlier what events can be listened to ?

Thanks — good follow-up. You asked which events you can listen to to detect speech earlier. Below are the events and signals you can use (from earliest to later): • START_OF_SPEECH ◦ Emitted by the STT node as soon as voice activity is detected and a speech segment begins. This is the earliest low-latency signal that someone has started speaking. • user_state_changed (new_state == "speaking") ◦ Emitted when the system’s VAD or turn-detector model marks the user as speaking. Useful when you rely on local VAD or the platform’s turn detection. • Interim transcription / INTERIM_TRANSCRIPT events ◦ Streaming STT models emit partial (interim) transcripts while the user is speaking. These let you start processing content before the final transcript is ready. • user_input_transcribed / final transcription ◦ Emitted once the STT produces a (final) transcription for the user’s utterance. Useful if you need higher accuracy before acting. How to combine them for earlier detection • Use START_OF_SPEECH or

user_state_changed

for the fastest detection to kick off pre-processing (e.g., start an LLM warm-up or show a typing indicator). • Use interim transcripts to refine or begin intent extraction while the user continues speaking. • Final transcription or

user_input_transcribed

should trigger the steady-state response generation. If you tell me which SDK or pipeline you’re using (browser, Node, Python, or the composite pipeline), I can show a short code snippet showing where to hook these events. Sources: Turns overview | LiveKit Documentation | Agent session | LiveKit Documentation | Agent speech and audio | LiveKit Documentation | examples/voice_agents | livekit/livekit_composite | DeepWiki