Speaker Timelines

A speaker timeline shows which meeting participant the platform identified as speaking over time.

Recall.ai includes the timeline in the recording’s Participant Events artifact, which is generated by default. Speaker timelines are also used for speaker-timeline diarization.

Download a speaker timeline

1. Wait until the recording’s Participant Events artifact has a status of done.
2. Call Retrieve Bot.
3. Find the relevant recording in recordings[] and read:

media_shortcuts.participant_events.data.speaker_timeline_download_url

Fetch that URL with an HTTP GET. The URL is presigned, so no Authorization header is needed.

The downloaded JSON contains participant information and the start and end timestamps for each interval. This example shows the key fields:

[
  {
    "participant": {
      "id": 100,
      "name": "Alex"
    },
    "start_timestamp": {
      "relative": 2.5,
      "absolute": "2026-10-25T17:00:02.500Z"
    },
    "end_timestamp": {
      "relative": 8.0,
      "absolute": "2026-10-25T17:00:08.000Z"
    }
  }
]

See the complete speaker timeline schema for all fields.

Timestamps

Each timestamp includes:

• relative: seconds since recording began, based on the bot’s latest in_call_recording status change (e.g., relative: 2.5 means 2.5 seconds after the recording began).
• absolute: the event’s date and time in ISO 8601 format.

Receive speaker events in real time

Configure a Real-Time Endpoint and subscribe to:

• participant_events.speech_on
• participant_events.speech_off

Note that tthe speech_on indicates that audio was emitted from the participant's tile, which can include but isn't limited to speech (e.g., couging, sneezing, and background noise can cause a speech_on event to fire).

See Real-Time Event Payloads for the event format. The timing of these events varies by meeting platform, as described below.

Understand platform differences

Speaker timelines reflect the meeting platform’s active-speaker signals.

For diarization methods and platform requirements, see Diarization.

FAQs

Does screen-share audio generate speaker events?

No. Speaker events are generated for meeting participants; screen-share audio does not generate speaker-change events.

Why is there a speech_on event but no transcript text for that period?

A speech_on event indicates that the meeting platform identified a participant as an active speaker. It does not guarantee that the audio contains recognizable words. Coughing, sneezing, or background noise can trigger an event without producing transcript text.

On platforms that report speaker changes, the active-speaker interval can also include silence before another participant starts speaking. Speaker events and transcript words therefore do not have a one-to-one relationship.

Review the recording to confirm that there was speech during a speech_on interval.