Developer documentation
Emotion and speaking style
Read how speech sounds, not only what was said: 15 emotion and 16 speaking-style scores from oruk, by file or stream, with or without a transcript.
Oruk scores the delivery of speech from the audio itself. The same words spoken two ways get different scores. There are 15 emotion labels and 16 speaking-style labels.
The labels
| Emotions | happy, excited, hopeful, sad, worried, angry, frustrated, disappointed, scared, disgusted, surprised, embarrassed, proud, relieved, neutral |
|---|---|
| Speaking styles | energetic, passionate, irritated, warm, playful, sarcastic, deadpan, hesitant, confident, sincere, skeptical, tired, formal, casual, impatient, distracted |
Choose a model
| Model | Returns | Transcript | Reference |
|---|---|---|---|
oruk-spectra-2 | All 31 scores for the whole clip | Yes, 25 languages | Spectra-2 |
oruk-resonance-2 | 31 scores and six signed axes for each range you commit | No | Resonance 2 |
oruk-resonance | Selected labels with timed segments | Yes, English | Quickstart |
oruk-realtime | Emotion scores for each phrase, live | Yes, live | Phrase emotions |
New organizations need separate approval for Resonance 2.
Score a clip
cURL
curl --fail-with-body https://speech-api.oruk.ai/v1/audio/analysis \
-H "Authorization: Bearer $ORUK_API_KEY" \
-F file=@sample.wav -F model=oruk-spectra-2The response carries emotions and styles as label and score pairs.
Reading the scores
- Each score runs from 0 to 1 on its own. Scores do not sum to one, and several can be high together.
- Read the labels returned. Do not hard-code a list; Realtime can return a different set from the file models.
- A score describes how speech sounds. It does not establish what a person feels, and should not be the sole basis for a decision about them.
- Transcription language coverage does not mean equal scoring accuracy in every language. Validate on your own recordings.
Related
Emotion recognition API overview · Benchmarks · Voice agents · Responsible use