When to use it
Recordings you already have
Calls, voicemails, uploads. Anything that is a file rather than a stream.
Captions and search
verbose_json gives you the timings to line text up with audio.More than one speaker?
Use audio intelligence — same endpoint,
speaker labels included.
Live microphone?
Use realtime. Chunking live audio into POSTs loses the
turn boundaries.
Request
Pass
language when you know it. Auto-detection is decided from the opening
seconds, so a recording that starts with noise, hold music, or a greeting in a
different language can be identified wrongly — and once it is, the whole
transcript is in the wrong language.
Timings and segments
verbose_json adds the detail you need to line a transcript up with audio:
Choosing a model
Two tiers, and the difference is whether you need to know who spoke.
Reach for
nur-stt-v1-pro when more than one person is talking — a meeting, an
interview, a two-sided call. It also reports whether each turn was on mic, off
mic, or too faint to trust, which is what lets you tell a bad recording from a
bad transcript. See Audio intelligence.
Read
GET /v1/audio/models rather than hardcoding an id. Speech tiers are
opt-in per environment, so a deployment only advertises what it can serve.Live audio
This endpoint is for audio you already have. For a microphone, use realtime — it streams, and it does its own end-of-turn detection. Do not chop live audio into repeated POSTs. It costs a round trip per chunk, and it cuts turns at arbitrary boundaries rather than where someone stopped talking, which is what makes the difference between a transcript you can read and a wall of fragments.Going further
Audio intelligence
Who said what, when — diarized long-form transcription.
Realtime
Live conversation, with turn detection.