GET /transcripts/{id}/result returns the conversation once the transcript is completed. Before that it returns 409.
What does the JSON look like?
What are speakers?
Each voice in the recording gets an id such as speaker_1 with its total speaking_ms. Ids are stable within one transcript and are not shared between transcripts. A recording can have up to 25 speakers.
What are turns?
A turn is one speaker talking without a break. It has the text, a start_ms and end_ms, and a words array with the time of every word. Times are milliseconds from the start of the recording.
What are overlaps?
An overlap is a stretch where two or more people talk at once. It has a start_ms, an end_ms and the turn_ids that overlap. Both speakers are transcribed in their own turns, so nothing is dropped.
What is activity?
activity lists when each speaker is audible, including short sounds with no words such as a laugh or a breath. Use it to draw a speaker timeline.
What are the analytics?
analytics summarizes the conversation: speech_ms, silence_ms, overlap_ms, switches_per_minute, median_response_ms, the list of silences and a breakdown per speaker.
How do I name the speakers?
Send PATCH /transcripts/{id} with speaker_names, for example {"speaker_names": {"speaker_1": "Ana"}}. The names appear on the transcript, in exports and in generated add-on text. The identify speakers add-on can propose names from the conversation itself.
Where are the add-on results?
Add-on output is in the result under addons, keyed by the add-on name. See add-ons.Last modified on October 26, 2026
