Developer documentation
Migrate from Deepgram
Move a Deepgram transcription integration to oruk: the endpoint, authentication and response fields that change, which model to pick, and what oruk does not replace.
Most of a transcription integration carries over. Three things change: how the audio is sent, the authorization header, and where the transcript sits in the response. Oruk also returns scores for how the speech sounded, which you can ignore or use.
The request
curl --fail-with-body "https://api.deepgram.com/v1/listen" \
-H "Authorization: Token $DEEPGRAM_API_KEY" \
-H "Content-Type: audio/wav" \
--data-binary @sample.wav
# transcript: .results.channels[0].alternatives[0].transcriptcurl --fail-with-body https://speech-api.oruk.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $ORUK_API_KEY" \
-F file=@sample.wav -F model=oruk-resonance
# transcript: .textWhat changes
| Deepgram | Oruk | |
|---|---|---|
| Authorization | Token scheme | Bearer scheme |
| Audio | Raw body or a URL | Multipart form field file |
| Model choice | Query parameter | Form field model |
| Transcript | Nested under channels and alternatives | Top-level text |
| Tone | Sentiment from the transcript text | Emotion and style scores from the audio |
| Live audio | WebSocket | WebSocket, oruk-realtime, 32 locales |
The Deepgram column describes its public v1/listen API in general terms. Check it against your own integration.
Pick the Oruk model
Long English recordings: oruk-resonance. Short clips in 25 languages: oruk-spectra-2. Low-cost short English clips: oruk-orukeet. Live audio: oruk-realtime. Compare them.
What Oruk does not replace
- Text-to-speech. Oruk does not generate speech.
- Fetching audio from a URL. Upload the file.
- Text features such as summaries and topic detection. Run those on the transcript with a language model.
- Languages outside each model’s list. Check coverage before you move traffic.
Before you cut over
Run both services on the same recordings and compare word error rate on your own audio. Published benchmarks do not predict accuracy on your microphones, accents and vocabulary.