Languages
Languages we can transcribe
Two: Russian and English. Our own engine takes those and nothing else — set the language yourself or let the service detect it. Below is every language the site covers, with the code of each, so you can check yours before you send a file.
Drop an audio or video file
or — mp3, wav, m4a, mp4, up to 64 MB and two hours
Transcription runs without an account: sign in to keep the text, the history and the exports.
What the engine takes
Two languages, and nothing is promised beyond them
| Language | Code | Automatic detection | What comes back |
|---|---|---|---|
| Russian | ru | yes | one segment per line, a timecode on every segment, speaker labels on request |
| English | en | yes | one segment per line, a timecode on every segment, speaker labels on request |
Set the language in the upload panel above: Detect automatically, Russian or English. There is no fourth value, and that is on purpose — a transcription service that accepts any language and quietly returns a text in the wrong one is worse than one that refuses.
The API takes the same three values in the language field of POST /v1/transcripts — ru, en or auto — and answers with an error to anything else, de included. That check lives in the code of the service, not in a promise on a page.
With auto the service asks the engine what it hears, tries three times when the engine is busy, and falls back to Russian if there is still no answer. Every transcript says where the language came from: user, auto or fallback. Detecting a language took 0.39 s in our measurement of 21 October 2026.
Sources: the service contract and the code of the service — apps/api/app/routers/transcripts.py (the three accepted values), apps/api/app/asr.py (the gate, the detection and the fallback), docs/CONTRACT.md. The formats are in apps/api/app/exports.py.
The full matrix
Every language the site covers
The site has a page about each of these languages. Two of them our engine transcribes; the rest we do not, and those pages say so and name the services that do. Search by a name or a code, narrow the list by a region or by an alphabet, sort it by any column.
Loading the list of the languages…
The table is not written by hand: apps/web/tools/languages/build_languages.py reads the code of the server, the fact base of the site and the catalog of the pages, and writes the file assets/languages/data/languages.json. The codes are the ones of ISO 639-1 and of BCP-47, taken from the IANA Language Subtag Registry. The region and the alphabet are a convenience for the filters, not a part of the contract.
Two different things
The language of the interface is not the language of the recording
This page is in English, and so are most of the pages of the site. That is the language of the interface: the language you read us in. The language of the recording is a separate thing, and it is the one our engine answers in — Russian or English, whatever language you read the site in.
The site publishes a page for many more languages than we transcribe: 63 of the languages in the table above are not supported by our engine. Those pages do not pretend otherwise. They say that we do not transcribe the language, they explain what our engine does instead with it, and they name the services that do support it — so you can go where your language is supported instead of guessing and paying for a text in the wrong language.
If the language you need is not here, write to us and name it: support@transcrib.si. The request goes to the people who build the engine, and it is the only way a language gets added.
Pages about a language
Where to go from here
- Transcribe Russian audio to text — the language the engine was built for, with the mixed Russian-English speech it was tuned on.
- Transcribe English audio to text — the second language of the engine, and the one that reaches the cloud fallback when it is enabled on the server.
- Transcribe German audio to text — a page we publish and a language we do not transcribe.
- Transcribe Spanish audio to text — the same: the page is here, the engine does not take the language.
The pages of the languages live in the table above: every row has a link to its page.
Questions
Questions people ask about the languages
How many languages can Transcrib Si transcribe?
Two: Russian and English. The upload panel offers exactly these two and the automatic detection, and the API accepts the same three values — ru, en and auto — and nothing else. We would rather refuse a file than return a text in a language nobody asked for.
How do I set the language of a recording?
In the upload panel above: Detect automatically, Russian or English. If you drive the service from your own code, send the same values in the language field of POST /v1/transcripts. Any other value is rejected with an error.
Does it detect the language on its own?
Yes, when the language is left on Detect automatically. The service asks the engine, tries three times when the engine is busy, and falls back to Russian if there is still no answer. Every transcript says which of the three happened — user, auto or fallback — so you always know what the engine heard.
What happens if my language is not supported?
Nothing silent. The file is not transcribed by our engine, and the page of your language says so plainly and names the services that do support it. If you need the language and it is not on the site, write to us and name it: support@transcrib.si.
Do I need an account to try it?
No. Transcription runs without an account: the free plan is $0, 3 files a day, up to 30 minutes and 64 MB per file, and no card. Sign in to keep the text, the history and the exports.
Try it on a recording from this week.
Russian or English, up to two hours and 64 MB, with a timestamp on every segment. The text comes out in six formats, and the first three files a day cost nothing.
Publisher: Transcrib Si. Author: Mark Burmistrov. Updated 2026-10-24.