Speaker Timelines

A speaker timeline is a history of active speaker events for a bot.

Recall uses this timeline for transcript diarization, and also exposes it through the Participant Events resource associated with the recording.

📘

Receiving the speaker timeline in real-time

If you have a real-time use case that requires the speaker timeline, you can configure a Real-Time Endpoint subscribed to the speech_on & speech_off events.

Platform differences

Meeting platforms have slightly different behavior for emitting active speaker events.

Zoom:

• Active speaker events are produced based on the underlying audio of the participant.
• This means that Zoom bots can receive active speaker events more frequently, even for the same user in the same monologue.

All other platforms:

• Other platforms do not send active speaker events based on the underlying audio. Instead, active speaker events are received when a new participant begins speaking.
• If receiving speech_off events using a Real-Time Endpoint, these events will be sent accordingly upon a new speaker transition, which also produces a speech_on event.

Timestamps

The timestamp for each active speaker events is offset in seconds relative to the bot's in_call_recording event.

Fetching the speaker timeline

After calling Retrieve Bot, you can download a speaker timeline from the media_shortcuts.participant_events.data.speaker_timeline_download_url field in the recordings object(s). View the response format response format here.

FAQs

Does screenshare audio produce speaker change events?

No, screenshares don’t emit speaker change events since these events are only generated from participants.