Response

Response bodyCopy Anchor Link

A successful request returns 200 with:

{
  "transcript": "oui c'est bien une Peugeot 3008",
  "turnAudioDuration": 4.7
}

• transcript: the transcript of the audio.
• turnAudioDuration: duration of the submitted audio in seconds.

These are the same two fields the Real-time API returns on turn_ended, minus the session and turn identifiers, which a single request does not need.

Anything else is an error; see Errors.

LimitsCopy Anchor Link

• Audio size: 48 MiB for the audio part. Larger parts are refused with 413.
• Duration: a request still running after 480 seconds of wall clock is abandoned and answered with 504.

The two are independent, so audio under the size limit can still hit the wall-clock one. Split long audio into turns and send one request each, which also gives the engine the per-turn context it is tuned for.