A pipeline for recordings.
An editorial team uploads long recordings and wants text with timestamps. What matters here is that no job starts twice and nothing is left lying around after retrieval.
How it is built.
-
Background from 24 MiB or 10 minutes
The pipeline submits larger files with
Audiobackground=true, up to 200 MiB and 28,800 seconds per file, within the limits of the model. -
No duplicate job
With a unique
Idempotency-Keyper recording, a retry does not create a second job. -
Poll the status
There is no webhook for audio jobs, so the pipeline polls
GET /v1/audio/transcriptions/{id}untilstatusiscompleted. -
Clean up
After retrieval, it deletes the job with
DELETE. After 24 hours at the latest, the result is gone anyway.
curl https://api.flowmodels.one/v1/audio/transcriptions \
-H "Authorization: Bearer $FLOWMODELS_API_KEY" \
-H "Idempotency-Key: recording-2026-11-09-001" \
-F 'model=groq/whisper-large-v3-turbo' \
-F 'language=en' \
-F 'timestamps=true' \
-F 'background=true' \
-F 'file=@recording.wav'
GET https://api.flowmodels.one/v1/audio/transcriptions/018f5678-1234-7abc-9def-0123456789ab
Authorization: Bearer $FLOWMODELS_API_KEY