Transcribe Audio

POST /openai/v1/audio/transcriptions

Transcribe audio into Bambara text.

Mirrors POST /v1/audio/transcriptions, so client.audio.transcriptions.create() works unchanged.

Parameters:

file: the audio file to transcribe.
model: sunjata-1 (default), djelia-asr-v2 or djelia-asr-v1.
response_format: json (default), text, verbose_json, srt, vtt. Djelia's ASR returns per-segment timestamps, so verbose_json, srt and vtt all carry real timings.
stream: when true, emits transcript.text.segment and transcript.text.delta events over SSE.

temperature reaches sunjata-1, which decodes with it; the older models ignore it.

Ignored: language (Djelia's ASR is Bambara-only), prompt, timestamp_granularities (segment timings are always returned in verbose_json).

Example Response (response_format="json"):

{"text": "Aw ni ce i ka kene wa."}

Request

The transcript, in the requested response_format. stream=true returns transcript.text.delta events instead.