Batch Stt Rest

Operator submits POST https://api.deepgram.com/v1/listen?model=nova-3 with Content-Type: application/json and body {"url":"https://media.example.com/call.wav"}.

Use JSON body with {"url":...} for URL ingestion. Deepgram fetches the media server-side; the URL must be publicly fetchable (or signed). Verify the media server returns audio/* Content-Type and a reasonable Content-Length. Treat non-2xx fetch errors as a Deepgram 400 with details in body.

Operator transcribes a 45-minute clinical interview and picks model=nova-3-medical.

Pick model=nova-3-medical for medical-domain audio (clinical vocabulary boost); pin the version. For general business audio use nova-3-general. Do not switch model per retry; benchmark on a held-out fixture before changing the default model in production.

Operator submits batch with callback=https://ops.example.com/dg-hook to avoid long-polling.

Request returns immediately with a request_id; transcript is POSTed to the callback URL when ready. Implement the receiver as idempotent on request_id — Deepgram retries the callback on non-2xx responses [REQUIRES-VERIFICATION on retry count and backoff]. Sign or scope the receiver to defend agains…