Speech-to-Text API with Emotion Detection

A dated transcription evaluation

The July 2026 Resonance 1 checkpoint recorded 4.44% average English word error rate across four public test sets (LibriSpeech clean/other · Common Voice · TED-LIUM). This historical result does not measure subsequent API builds or predict accuracy on your recordings. The three rows below were measured in that panel; lower is better.

Open ASR A and B are anonymized checkpoints; public reproducible identifiers are unavailable. The release contains aggregate WER without confidence intervals, so the 0.03-point difference does not establish a significant lead. Read the full transcription results and source limits.

Fast transcription, included in your plan

The hosted Orukeet preview transcribes English audio at $0.00045 per minute. Hobby is $9/month with up to 20,000 transcription minutes plus 250 other-model minutes; optional emotion detection and speaker diarization use the Orukeet allowance. Use your Oruk API key with recordings up to 60 seconds and 4 MiB per request.

Need local multilingual transcription?

Orukeet is our downloadable recognizer for 25 languages, with native, ONNX and NeMo formats. It runs on your computer and does not include the emotion or speaking-style outputs of this hosted API. Install it and transcribe a supplied recording or inspect its paired ASR evaluation. Those model results are separate from the hosted API’s historical WER table above.

What a plan minute costs in context

Published rates for prerecorded transcription, per audio minute. The oruk row is the effective rate on the Builder plan when its included minutes are used:

Compare your expected total bill and required outputs. When you need emotion and speaking style alongside words, the analysis endpoint returns them together. Builder includes 2,500 minutes a month, about 41.7 hours of Resonance transcription, and every plan starts with a 7-day free trial.

The transcript is half the signal

“Great, thanks.” transcribes identically whether the delivery sounds warm or sarcastic. A plain transcription response stops at the words. Use /v1/audio/analysis and the response adds selected labels from 15 emotion and 16 speaking-style categories, with scores and an inline-tagged transcript:

"text":        "great, thanks",
"tagged_text": "great, thanks <|frustrated|> <|sarcastic|>",
"emotions":    [{ "label": "frustrated", "score": 0.83 }],
"styles":      [{ "label": "sarcastic",  "score": 0.78 }]

A single Resonance unified analysis request uses one plan minute per minute of audio, including its transcript, emotion, and speaking style. Separate transcription and analysis requests each consume plan minutes. Choose analysis first when you need all three outputs. Plan allowances and overage are on the pricing page.

The response excerpt above is illustrative. For a complete request and score interpretation, follow the Python recording walkthrough. Use the speaker diarization workflow when speaker turns are part of your integration.

Exactly what you are buying

The limits below are stated here so you do not discover them after integrating.

Prerecorded files, not live captions

Resonance transcribes recorded English audio. For live captions, use the separate Realtime preview: 32 locales, automatic language detection, and phrase-level emotion scores.

Input

Resonance 1 accepts prerecorded English audio: WAV, FLAC, MP3, M4A, OGG, or WebM, up to 30 MB and 60 minutes per file.

Realtime preview

The separate Realtime preview streams transcription in 32 locales with automatic language detection and phrase-level emotion scores. Resonance 2 returns 15 emotion and 16 speaking-style scores for explicit WebSocket commits or file requests, without transcription or timestamps. Original unified analysis uses the English file endpoints.

File limits and formats

Up to 30 MB and 60 minutes per file. WAV, FLAC, MP3, M4A, OGG, and WebM are supported.

Request behavior

File endpoints return a synchronous JSON response with measured audio duration. Realtime uses WebSocket events for live tokens and phrase-level emotion. Resonance 2 WebSocket commits return expression scores for a buffered sample range.

Pricing

Monthly plans start at $9. Resonance, Fourier, Realtime, and Proficiency share one audio-minute allowance. Orukeet has a separate allowance; its optional tasks use that allowance at their listed rates. The other-model allowance ranges from 250 minutes on Hobby to 20,000 on Production. Extra usage follows each allowance's published rates. Self-serve plans start with a 7-day free trial. Contact us for Enterprise pricing. Details on the pricing page.

Data retention and training

Request audio is discarded after processing; Resonance 2 streams also buffer audio in memory until reset or session closure. Resonance 2 result payloads, without raw audio, are stored for durable settlement and same-ID replay. Authorized replay is available for 24 hours; this is not a physical-deletion deadline. Optional speaker diarization retains uploaded audio for up to 48 hours and speaker-label results for up to 24 hours. Oruk does not use customer audio or outputs to train, fine-tune, or evaluate models without your explicit written agreement.

What scores are — and are not

Emotion scores are model annotations selected with model-specific thresholds, not probabilities of inner feelings. They are not intent, truth, medical, employment, or psychological judgments, and must not be the sole basis for consequential decisions. See responsible use.

Model and version stability

Models are versioned; Resonance and Fourier are stable, Realtime and Proficiency 1 are in preview. Changes are announced on the changelog.

Support

Email access@oruk.ai; the production guide covers retries, limits, and error handling.