Speech-to-text reference

Details that apply across the transcription API rather than to one transport.

Timestamps. Word and segment start and end times in the OpenAI verbose_json shape, plus rendered srt and vtt caption files. Use them to sync captions to playback or jump to a moment in a recording.

Diarization. Speaker attribution for interviews, meetings, and calls, plus how to read a word that carries no speaker.

Languages. The thirty languages Rhythm transcribes, the narrower set Symphony holds a conversation in, and how the language field behaves.

Timestamps and diarization ride the verbose REST response, so request both with response_format=verbose_json. REST has the request shape.