A5S Async
Every word, and who said it.
Our most accurate model for recorded audio. Meetings, calls, interviews and media come back as transcripts with speaker labels, even when people talk over each other.
Audio: AMI Meeting Corpus, CC BY 4.0. Model output as returned, lowercase and verbatim.
Hear the difference.
Real meetings from public benchmarks. Press play, choose another provider, and follow both transcripts word by word against the human reference.
Loading the listening demo…
Built for real conversations.
Speaker labels, timestamps and subtitles on every transcript, with nothing extra to buy.
Speaker labels, included
Every utterance and every word says who spoke it: speaker_0, speaker_1 and so on.
Holds up under crosstalk
Where half or more of the speech overlaps, it makes 37% fewer errors than the next-best model.
Timestamps and subtitles
Start and end times for every word, and SRT or VTT captions from the same transcript.
Names, not numbers
Enroll the people you expect to hear and transcripts come back with their names, through AirCaps Speaker ID.
Key terms
Bias recognition toward the names, products and jargon in your audio with a list of key terms per request.
Long files, simple jobs
Recordings up to 10 hours by URL or upload. Poll for the result, or get a webhook when it is ready.
Ranked first on real meetings.
cpWER averaged over NOTSOFAR-1, AMI, CHiME-6, AISHELL-4 and AliMeeting: 29% fewer errors than the next-best model.
Where it fits.
Notetakers and meeting assistants
Attribute every decision and action item to the person who said it, even in loud, fast meetings.
Captions and subtitles
Word-timed transcripts and SRT or VTT files for interviews, podcasts and video, with each speaker marked.
Calls, interviews and research
Turn recorded conversations into searchable text that says who said what, ready for analysis and review.
Transcribe a recording in a few lines.
- Send a URL or upload a file, up to 10 hours
- Poll for the result, or pass a
webhook_url - Utterances and words with speakers and timestamps
- Subtitles as SRT or VTT from the same transcript
import os, time, requests
API = "https://api.aircaps.com/v1"
H = {"Authorization": f"Bearer {os.environ['AIRCAPS_API_KEY']}"}
job = requests.post(f"{API}/transcripts", headers=H, json={
"model": "a5s-async-v1",
"audio_url": "https://example.com/meeting.mp3",
}).json()
while job["status"] in ("queued", "processing"):
time.sleep(3)
job = requests.get(f"{API}/transcripts/{job['id']}", headers=H).json()
for u in job["utterances"]:
print(f"[{u['start']:.1f}s] {u['speaker']}: {u['text']}")curl https://api.aircaps.com/v1/transcripts \
-H "Authorization: Bearer $AIRCAPS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "a5s-async-v1",
"audio_url": "https://example.com/meeting.mp3",
"webhook_url": "https://your.app/hooks/aircaps"}'
# → {"id": "tr_01k6z...", "status": "queued", ...}
# the webhook fires when the transcript is ready
curl https://api.aircaps.com/v1/transcripts/tr_01k6z.../subtitles?format=srt \
-H "Authorization: Bearer $AIRCAPS_API_KEY"Specifications.
Everything else is in the model reference and limits.
- Model ID
a5s-async-v1- Interface
- REST, asynchronous jobs, webhooks
- Input
- Any common audio or video format, by URL or upload
- Length
- 0.1 s to 10 hours, up to 5 GB per file
- Output
- Utterances and words with speaker labels and timestamps; SRT and VTT subtitles
- Language
- English, with regional codes such as en-US, en-GB and en-IN
- Speaker names
- From enrolled voices, with AirCaps Speaker ID enabled
- Price
- $0.18 per audio hour, billed per millisecond
Questions.
Are speaker labels extra?
No. Every A5S Async transcript includes speaker labels, word timestamps and subtitles at $0.18 per audio hour.
How accurate is A5S Async?
It makes 29% fewer errors than the next-best model across five public meeting benchmarks, with the lowest error on every one. The research note has every result.
How fast is it?
Through the API, an hour-long meeting comes back in 9.8 seconds (median), from upload to finished transcript.
What does the text look like?
Verbatim and lowercase, without punctuation. Disfluencies such as “uh” and “um” are kept.
Which languages does it support?
English, including regional codes such as en-US, en-GB and en-IN that match the speakers' accent.
Do you train on my audio?
No. See Data and privacy in the docs.
More models.
Every word, and who said it.
Run A5S Async on your own recordings. Every account includes 5 free hours.