API
Everything the site does, from code. Send audio files or links to them, get one transcript back per file. Same credits, same prices: one credit per started hour.
Start here
- Base URL
https://audiomenti.com/api/v1. JSON in, JSON out. CORS is open. - Authenticate with
Authorization: Bearer am_...on every call. - A job is a set of files transcribed together and billed together. It goes
draft,queued,running, thendoneorfailed. - Machine-readable: /openapi.json, /llms.txt, /agents.md.
Getting a key
A key is made by verifying an email address, on the login page or with
two calls. Sending agree: true accepts the terms and the
privacy policy. The key is free; credits are what cost money.
curl -X POST https://audiomenti.com/api/v1/register/start \ -H "Content-Type: application/json" \ -d '{"email": "you@example.com", "agree": true}' # open the link in the email, or send its token: curl -X POST https://audiomenti.com/api/v1/register/confirm \ -H "Content-Type: application/json" \ -d '{"token": "<from the emailed link>"}' # -> {"key": "am_..."} # every call after that: AUTH="Authorization: Bearer am_YOUR_KEY" curl https://audiomenti.com/api/v1/me -H "$AUTH"
Transcribe files from URLs
The shortest way: give us links we can download without logging in. With
start: true the job is paid from your credits and started in the same call.
Up to 200 files by URL per job.
curl -s -X POST https://audiomenti.com/api/v1/jobs -H "$AUTH" \ -H "Content-Type: application/json" \ -d '{ "files": [ {"url": "https://example.com/audio/interview-anna.mp3"}, {"url": "https://example.com/audio/interview-jakob.mp3", "path": "interviews/jakob.mp3"} ], "language": "en", "format": "srt", "webhook_url": "https://example.com/hooks/audiomenti", "start": true }'
The answer, and what GET /api/v1/jobs/:id returns from then on:
{
"id": "job_4f0c2a9d81b7e6a35c1d9e02",
"status": "running",
"language": "en",
"format": "srt",
"files_total": 2,
"files_done": 1,
"files_failed": 0,
"estimated_seconds_of_audio": 5310,
"seconds_of_audio": null,
"credits_quoted": 2,
"credits_charged": 2,
"credits_owed": 0,
"estimated_wait_seconds": 170,
"url": "https://audiomenti.com/job/job_4f0c2a9d81b7e6a35c1d9e02",
"download": null,
"files": [
{"n": 1, "path": "interview-anna.mp3", "status": "done", "seconds": 2655, "words": 6120, "transcript": null},
{"n": 2, "path": "interviews/jakob.mp3", "status": "running", "seconds": null, "words": null, "transcript": null}
]
} seconds is optional per file. If you leave it out, the quote is estimated from
the file's size and settled against the real length when the job finishes: you are charged
for what the audio actually ran to, added together and rounded up to whole hours.
Upload files
For files that are not on the web: create the job, send the bytes, start it. Each file can be up to 10 hours long, and a job up to 900 files.
# 1. say what is coming: name, size in bytes, length in seconds if you know it curl -s -X POST https://audiomenti.com/api/v1/jobs -H "$AUTH" \ -H "Content-Type: application/json" \ -d '{"files": [{"path": "board-meeting.m4a", "bytes": 48211034, "seconds": 3412}], "language": "da", "format": "txt"}' # -> {"id": "job_...", "status": "draft", "credits_quoted": 1, ...} # 2. send the bytes (files are numbered from 1, in the order you listed them) curl -s -X PUT https://audiomenti.com/api/v1/jobs/$JOB/files/1 -H "$AUTH" \ --data-binary @board-meeting.m4a # 3. pay from your credits and start curl -s -X POST https://audiomenti.com/api/v1/jobs/$JOB/start -H "$AUTH"
Larger files go up in parts:
# files over 95 MB go up in parts curl -s -X POST https://audiomenti.com/api/v1/jobs/$JOB/files/1/multipart -H "$AUTH" # -> {"n": 1, "part_size": 33554432, "parts": 12} # each part is part_size bytes, the last one whatever is left curl -s -X PUT https://audiomenti.com/api/v1/jobs/$JOB/files/1/parts/1 -H "$AUTH" --data-binary @part-01 # -> {"part": 1, "etag": "..."} curl -s -X POST https://audiomenti.com/api/v1/jobs/$JOB/files/1/complete -H "$AUTH" \ -H "Content-Type: application/json" \ -d '{"parts": [{"part": 1, "etag": "..."}, {"part": 2, "etag": "..."}]}'
A draft that is not started within 24 hours is deleted along with whatever was uploaded to it.
Get the text, delete the job
# poll until status is "done" (or take the webhook) curl -s https://audiomenti.com/api/v1/jobs/$JOB -H "$AUTH" # one transcript, in any format curl -s "https://audiomenti.com/api/v1/jobs/$JOB/files/1/transcript?format=json" -H "$AUTH" # everything: one file for a one-file job, otherwise a zip laid out like what you sent curl -s -OJ "https://audiomenti.com/api/v1/jobs/$JOB/download?format=srt" -H "$AUTH" # wipe it: audio still waiting, every transcript, the job itself curl -s -X DELETE https://audiomenti.com/api/v1/jobs/$JOB -H "$AUTH"
Downloads are free and repeatable in any format. language is one of
25 codes from the languages page
(GET /api/v1/languages); the model detects the spoken language itself, so the
field is a note for you rather than a switch.
Output formats
| format | what you get |
|---|---|
txt | Plain text (.txt). The words in paragraphs. For reading, quoting and pasting into a document. |
timestamps | Text with times (.txt). Each paragraph starts with the time it was said, so you can find it in the recording. |
md | Markdown (.md). Paragraphs with a time heading every few minutes. For notes apps and wikis. |
srt | SRT subtitles (.srt). Numbered captions with times. For video editors, YouTube and most players. |
vtt | WebVTT subtitles (.vtt). Captions for the web: the format browsers play next to a video. |
json | JSON (.json). Every segment and every word with its time. For code. |
Credits and prices
One credit is up to one hour of audio. A job costs one credit per started hour of all its files added together. Ask for a quote without a key:
# no key needed: what would 37 files, 61 hours in all, cost and how long would it take? curl -s "https://audiomenti.com/api/v1/pricing?seconds=219600&files=37" # -> "quote": {"credits": 61, "usd": 19, "estimated_wait": "about 14 minutes", ...}
Credits come from a monthly plan or a one-time purchase: 1 h for $1.49, 5 h for $3.49, 10 h for $5, 50 h for $15, 100 h for $25, 200 h for $45, 500 h for $100, 1000 h for $190, 2000 h for $360, 5000 h for $850, or any whole number of hours at the rate for that size. All prices. To buy from code, ask for a checkout link and open it:
curl -s -X POST https://audiomenti.com/api/v1/checkout -H "$AUTH" \ -H "Content-Type: application/json" \ -d '{"hours": 250}' # -> {"url": "https://checkout.stripe.com/...", "credits": 250}: open url in a browser to pay
If a job turns out longer than quoted and the balance cannot cover the difference, it
finishes with credits_owed above zero and its downloads answer
402 until you add credits and call /start again. A file that
fails costs nothing.
Webhooks
Give a job a webhook_url (https) and we POST to it once, when the job is done
or has failed:
{
"event": "job.done",
"job": { "id": "job_...", "status": "done", "download": "https://audiomenti.com/api/v1/jobs/job_.../download", "files": [ ... ] }
} Each delivery is signed, the way Stripe signs:
# X-Audiomenti-Signature: t=1763380000,v1=5f2c... # v1 = hex( HMAC-SHA256( webhook_secret, t + "." + raw_body ) ) # webhook_secret is on GET /api/v1/me
A delivery is tried once and times out after 8 seconds, so treat it as a nudge and read the
job with GET when it arrives. Polling every 10 to 30 seconds works just as well.
Errors and limits
Every error has the same shape, with a code to switch on and a message to read:
{
"error": {
"code": "insufficient_credits",
"message": "This job needs 61 credits and the account has 12."
}
} | status | meaning |
|---|---|
| 400 | Something in the request is wrong; the message says what. |
| 401 | missing_key or invalid_key. |
| 402 | insufficient_credits: buy more, then start again. needed says how many. |
| 404 | No such job or file, or it belongs to another account. |
| 409 | Not in the right state: not_uploaded, already_uploaded, already_paid. |
| 429 | rate_limited: the daily request allowance is used up. |
Each call with a key counts as one request against your plan's daily allowance: Free 1,000, Plus 5,000, Pro 20,000, Business 50,000, Studio 100,000, Network 200,000 a
day, reset at midnight UTC. X-RateLimit-Remaining on every answer says what is
left.
All endpoints
POST | /api/v1/register/start Email a login link. Body: email, agree. |
POST | /api/v1/register/confirm Trade the emailed token for the key. |
GET | /api/v1/me Balance, plan, requests today, webhook secret, recent ledger. |
PATCH | /api/v1/me Body: notify (email me when a job is ready). |
DELETE | /api/v1/me Body: confirm "delete". Wipes the account and every transcript. |
POST | /api/v1/key/rotate A new key. The old one stops working at once. |
GET | /api/v1/pricing Prices and limits. With seconds and files: a quote. |
GET | /api/v1/languages The languages, with error rates. |
POST | /api/v1/jobs Create a job from files (by URL, or to be uploaded). |
GET | /api/v1/jobs Your jobs, newest first. limit up to 200. |
GET | /api/v1/jobs/:id One job and its files. |
PATCH | /api/v1/jobs/:id Body: format (the default for its downloads). |
DELETE | /api/v1/jobs/:id Wipe the job. Unfinished work is refunded. |
PUT | /api/v1/jobs/:id/files/:n Upload file n, up to 95 MB. |
POST | /api/v1/jobs/:id/files/:n/multipart Begin a multipart upload. |
PUT | /api/v1/jobs/:id/files/:n/parts/:part Upload one part. |
POST | /api/v1/jobs/:id/files/:n/complete Finish a multipart upload. |
POST | /api/v1/jobs/:id/start Pay from credits and start. |
GET | /api/v1/jobs/:id/files/:n/transcript One transcript. format, inline=1. |
GET | /api/v1/jobs/:id/download All transcripts. format. |
POST | /api/v1/checkout A Stripe checkout link: job, pack, hours or plan. |
POST | /api/v1/billing/portal A link to the Stripe billing portal. |
What happens to the audio
The same as on the site. Audio is stored only until each file is transcribed, then deleted. The transcription runs on a GPU rented from Replicate, a third party in the United States, which removes its copy within an hour. Transcripts stay until you delete the job. If you build a product on this API, tell your own users. Privacy policy.
Stuck, or need more than the limits allow? contact@audiomenti.com.