REST API

Base URL: https://api.speechrouter.ai. Errors use an OpenAI-compatible envelope with a closed set of codes:

{ "error": { "code": "model_not_found", "message": "…", "type": "invalid_request_error" } }

The machine-readable spec: openapi.yaml.

POST /v1/audio/transcriptions

Batch transcription. Multipart form; Bearer auth. Fields, formats, and limits: Batch transcription.

POST /v1/tokens

Mint a short-lived client token. Bearer auth with a full API key (tokens cannot mint tokens).

Body field	Default	Range
ttl_seconds	60	10–300

Returns { "token": "st_…", "expires_at": "…", "ttl_seconds": 60 }.

GET /v1/models

The live catalog — no auth required. Each entry carries the slug, display name, modes (streaming / batch), capabilities, and pricing in the vendor’s native unit:

{
  "slug": "soniox/stt-rt-v5",
  "modes": ["streaming"],
  "pricing": { "per_session_hour_usd": 0.12 },
  "capabilities": { "diarization": true, "word_timestamps": true, "...": "..." }
}

The catalog is the billing engine’s own source of truth — what it says is what you pay.

POST /v1/listen

Deepgram-compatible prerecorded transcription (Token auth scheme) — see Deepgram compatibility.

GET /up

Health probe. 200 when the gateway is serving.