Developer documentation
Speech-to-text
Choose oruk transcription: English recordings with Resonance 1, 25 languages with Spectra-2, short English clips with Orukeet, or live audio with Realtime.
Four Oruk models transcribe speech. They differ in language coverage, clip length and whether words arrive live. Pick one, then follow its reference.
Choose a model
| Model | Use it for | Languages | Length | Reference |
|---|---|---|---|---|
oruk-resonance | English recordings, with emotion, style and timed segments | English | Up to 60 minutes, 30 MB | Quickstart |
oruk-spectra-2 | Short multilingual clips, with emotion and style scores | 25 | Up to 60 seconds | Spectra-2 |
oruk-orukeet | Low-cost English text at $0.00045 a minute | English | Up to 60 seconds | Orukeet |
oruk-realtime | Live words over WebSocket, with an emotion score for each phrase | 32 locales | Up to 10 minutes a session | Realtime |
Transcribe a file
Send the audio as multipart form data with a Bearer key. The response has text and usage; other fields depend on the model.
curl --fail-with-body https://speech-api.oruk.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $ORUK_API_KEY" \
-F file=@sample.wav -F model=oruk-resonanceTranscribe live audio
Realtime returns words as they are spoken. Hosted Orukeet and Spectra-2 also accept audio over a WebSocket, but return one final result after you commit or finish. See the three streaming paths compared.
Speaker labels
Resonance 1 and Realtime can label speakers within plan minutes. Hosted Orukeet offers it as a separately priced task. Spectra-2 does not label speakers.
Run it locally
Orukeet is also a downloadable model for 25 languages. Install it and transcribe a recording.
Related
Speech-to-text API overview · Model comparison · SDKs · Migrate from Deepgram