API capabilities

Supported today

• Multilingual realtime speech. Realtime streams live transcript tokens and separate phrase-level emotion events over WebSocket. Transcription supports 32 locales with automatic language detection; that coverage does not establish emotion accuracy in every locale.

• File transcription. POST /v1/audio/transcriptions supports 25-language transcription with Spectra-2. Original Resonance 1, Fourier and hosted Orukeet transcribe English. Input limits and response fields depend on the selected model.

• Multilabel emotion detection. Original Resonance 1 and Fourier return threshold-selected emotion labels with a highest-scoring fallback. Spectra-2 returns all 15 emotion scores for the whole clip. Resonance 2 returns continuous scores and selected labels; its selected labels can be empty.

• Speaking-style classification. Original Resonance 1 and Fourier return threshold-selected speaking-style labels; an empty style array is valid. Spectra-2 returns all 16 style scores for the whole clip. Resonance 2 returns continuous scores and selected labels without transcription.

• Combined affect. Original Resonance 1/Fourier affect requests and the dedicated Resonance 2 route return emotion and speaking-style analysis without transcript output.

• Unified analysis. Spectra-2 returns multilingual text and clip-level emotion and style scores in one request, without word timings, timed segments or tagged text. Original Resonance 1 and Fourier return English text with timed segments and tagged text.

Scope and limits

• Realtime returns live tokens and phrase-level emotion. Resonance 2 separately returns emotion and speaking-style scores for committed WebSocket audio ranges, without a transcript or timestamps. Original unified analysis remains a prerecorded file endpoint.

• Language support is model-specific: Spectra-2 file transcription supports 25 languages; Realtime transcription supports 32 locales. Original Resonance 1, Fourier and hosted Orukeet transcription remain English. Transcription coverage does not establish expression accuracy across those languages.

• No intent or deception detection. The sarcastic speaking-style label scores delivery — how speech sounds — not what the speaker meant.

• No claims about a speaker’s inner mental state — outputs are model annotations of how speech sounds.

Endpoints

POST /v1/audio/transcriptions

Returns a transcript. Spectra-2 supports 25 languages and clip-level label scores; Resonance 1 and Fourier return English transcripts with segment and word timings.

POST /v1/audio/emotions

Returns multilabel emotion scores per acoustic segment from Resonance 1 without running transcription. The response carries no transcript text. One minute of audio uses one plan minute, just like unified analysis.

POST /v1/audio/styles

Returns multilabel speaking-style scores per acoustic segment. The response intentionally omits transcript text.

POST /v1/audio/affect

Returns emotion and speaking-style scores without transcript text.

POST /v1/audio/analysis

Returns transcription, emotion, and speaking style. Spectra-2 returns multilingual text and clip-level scores; Resonance 1 and Fourier return English text with timed segments and tagged text.

POST /v1/audio/proficiency

Preview: continuous-pilot-v2 returns word and recording pronunciation scores on 0–1 with accuracy, fluency and prosody. Original CEFR fields and the 0–5 score remain compatible. Read continuous_check for pronunciation validity and the top-level check for CEFR validity and billing. Subscription accounts use plan minutes.

WS /v1/realtime

The Realtime preview accepts live PCM16 audio and returns transcript tokens and phrase-level emotion scores. Use the runnable streaming guide for session setup, events, and shutdown. Its phrase outputs differ from the full English file emotion and style analysis.

Models

Spectra-2Stable

Transcribe a recording and receive 15 emotion and 16 speaking-style scores in the same response, using your existing Oruk API key.

Resonance 2Preview

Our new affect model returns 31 continuous scores, with six signed axes for opposite labels and 19 independent scores. Available through the dedicated Resonance 2 API endpoint.

OrukeetPreview

Transcribe English audio with your Oruk API key. Add emotion detection or speaker diarization when your application needs them.

Resonance 1Stable

Transcribe English audio and analyze emotion and speaking style in one request, with results aligned to segments of the recording.

FourierStable

Transcribe English audio and score 15 emotion labels in parallel. The API also returns 16 speaking-style labels through the same style model used by Resonance 1.

RealtimePreview

Live transcription in 32 locales, with automatic language detection and a separate stream of phrase-level emotion scores. Speaking-style analysis uses the English file API.

ProficiencyPreview

Word and recording pronunciation scores on 0–1, with the original CEFR and fluency fields

Limits by model

Choose the model before preparing audio. File formats, duration limits and streaming protocols differ; the linked references give the full request contract. Resonance 2 requires separate organization approval.

Speech understanding models share a plan allowance. Orukeet has a separate allowance and task rates; its actual-duration metering differs from the other models' one-second minimum. See pricing and billing terms.