| Model | What it does |
|---|---|
A5S Async v1a5s-async-v1 | Recorded audio (0.1 s to 3 h) → a transcript with speaker labels, word timestamps and subtitles. REST, async jobs |
AirCaps Speaker ID-v1aircaps-speaker-id-v1 | A clip (0.5 s to 2 min) → a score for each of your enrolled speakers. REST, synchronous |
AirCaps Diarization-v1aircaps-diarization-v1 | Recorded audio (0.1 s to 3 h) → who spoke when: speaker segments as JSON or RTTM. REST, async jobs · enabled on request |
A5S Streaming v1a5s-streaming-v1 | Live audio in English and eight other languages → punctuated captions with speaker labels as people speak. WebSocket · early access, on request |
| A5S Small V2 Streaming | Compact streaming speech-to-text for noisy, far-field audio. VPC, on-premise, air-gapped and on-device deployments only · talk to sales |
A5S Async v1
a5s-async-v1 · REST · batch guide · benchmarks
- Recorded audio in any common format, from 0.1 s to 3 hours.
- Speaker labels (
speaker_0,speaker_1, …), word timestamps, utterances, SRT/VTT subtitles. - Optional key terms in English transcripts: up to 60 per transcript help it recognize names and jargon (
keyterms, see Key terms). - Optional speaker names from your enrolled voices (
speaker_identification, with AirCaps Speaker ID-v1, at no extra cost). See Speaker names. - English, Spanish, French, German, Portuguese, Hindi, Japanese, Korean and Chinese:
language_codeisen(or a regional English code such asen-US,en-GBoren-INthat matches the speakers’ accent),es,fr,de,pt,hi,ja,koorzh: set it to the language of the audio (recorded audio has no automatic detection). See Languages. - Punctuated, cased text with numbers as spoken (“twenty five dollars”); disfluencies (“uh”, “um”) are kept.
- Optional written forms for numbers, money, percentages, dates and times, in every supported language (
smart_format, see Smart formatting).
AirCaps Speaker ID-v1
aircaps-speaker-id-v1 · REST, synchronous · speaker ID guide · evaluation
- Recognize enrolled people by voice. Enroll each person with one or more clips, then send a clip and get back a score for every enrolled speaker, best first.
- Also names the speakers of A5S Async v1 transcripts (
speaker_identificationon create a transcript). - Synchronous: one clip per request, 0.5 s to 2 minutes; the scores are the response. No jobs, no polling.
- You set the threshold: scores are similarities from -1 to 1, with no accept/reject decision. The guide has measured score ranges to start from.
- The whole clip is compared as given. Crop clips to one person’s speech first (for example with AirCaps Diarization-v1 or a transcript’s utterances).
- Works from the voice, not the words: no
language_code. - Available to every account.
AirCaps Diarization-v1
aircaps-diarization-v1 · REST · diarization guide · benchmarks · enabled on request
- Who spoke when, with no transcript: speaker segments (
speaker_0,speaker_1, …) with millisecond times, overlapping speech included, plus each speaker’s speaking time. - Finds the number of speakers itself. Works from the voices, not the words: no
language_code. - Recorded audio in any common format, from 0.1 s to 3 hours. Results as JSON or RTTM.
- Enabled per account: talk to sales for access.
A5S Streaming v1
a5s-streaming-v1 · WebSocket · streaming guide · early access, on request
Early access. Available on request to select organizations. Talk to sales to request access.
- Live 16 kHz mono PCM16 audio in, partial and final text out as people speak; choose it with
"model": "a5s-streaming-v1"instart. - English, Spanish, French, German, Portuguese, Hindi, Japanese, Korean and Chinese.
- Punctuated, cased text with live speaker labels and each utterance’s language: always set
diarization: trueto receive them. - Optional written forms for English finals (
smart_format: true) and key terms in English sessions. - Not yet: per-word timestamps and speaker names.
- Sessions last up to 30 minutes (up to 1 hour on request).
A5S Small V2 Streaming
Deployments only · benchmarks A5S Small V2 Streaming is a compact streaming model for noisy, far-field audio. It is offered for VPC, on-premise, air-gapped and on-device deployments only. Talk to sales.- Live English speech-to-text: punctuated, cased text as people speak.
- Live speaker labels.
- Key terms to bias recognition toward names and jargon.
Choosing
Languages
- For English, a regional code (
en-US,en-GB,en-IN, …) that matches the speakers’ accent gives the best results. - A region on another language’s code is accepted and decodes as that language (
es-MXises,pt-BRispt). - Set the language of the audio: there is no automatic detection.
- A transcript’s
language_codeis the code you sent, without its region. GET /v1/modelslists each model’slanguages.
Access
Every account can use A5S Async v1 and AirCaps Speaker ID-v1. AirCaps Diarization-v1 is enabled per account on request; without it, its requests return403 model_access_denied. A5S Streaming v1 is in early access for select organizations; without access, a session that names it ends with model_access_denied (close code 4403). A5S Small V2 Streaming is only offered for VPC, on-premise, air-gapped and on-device deployments. GET /v1/account shows enabled for each model. For access, talk to sales.
GET /v1/models lists the models.
