Skip to main content
Use the id value as the model parameter in API requests. 5 models currently available.

​
Usage

Speech-to-text models transcribe spoken audio into written text. They are accessed via the Audio Transcriptions API.

​
Supported audio formats

wav, wave, flac, m4a, aac, mp4, mp3, ogg, oga, webm A file is accepted when either its MIME type or its extension is on this list, and it must also pass a magic-byte check on the uploaded bytes. Renaming a file to a supported extension is not sufficient.

​
Response formats

​
Timestamps

Set timestamps: true to receive timing data alongside the transcript. The response adds a timestamps object whose granularity depends on the model:
nvidia/parakeet-tdt-0.6b-v3 is the default model, and it accepts timestamps: true without returning timings. There is no error and no warning — the response simply contains text and nothing else. If you need timings, choose a model from the table above and check that the timestamps key is present before reading it.
Word entries are { "word": "...", "start": 0.0, "end": 0.5 } and segment entries are { "text": "...", "start": 0.0, "end": 3.2 }, with all times in seconds.
Pricing is billed per second of input audio. See the Audio Transcriptions API for request examples and parameter details.