How to detect speaker turn detection ?
Gladia provides speech_start and speech_end events so you can detect exactly when a person starts and stops talking, much faster than waiting for transcripts.
Why use them?
Ultra low latency: trigger in real time as soon as speech is detected.
Speaker turn detection: combine both events to know when a speaker takes or ends a turn.
Better UX: react instantly (show “speaking” state, cut latency, manage bot handoffs).
💡 Pro tip
When building voice agents, low latency is crucial. By combining speech activity detection with partial transcriptions, you can feed text to your LLM faster — which means it can start making sense of the conversation even before the final, polished transcription is ready.
How it works
speech_start → fired when human speech begins.
speech_end → fired when speech stops.
Transcript text might arrives later, but you already know the turn boundaries.