What Is Multilingual Speech Recognition?

What is multilingual speech recognition? It is a single system that recognises speech in many languages, instead of one separate model for each. The system learns shared patterns of sound and decides which language it is hearing. Quality still differs by language, and some are served far better than others.

Transcribe a recording

Send your voice message, audio or video to @potok_now_bot on Telegram and the text comes back in the chat.

Open @potok_now_bot in Telegram

Files up to 2 GB · export to TXT, MD, DOCX, PDF, SRT and VTT · speaker labels and timestamps · about 30 languages

Key facts

One model for many languages is trained on recordings in all of them, so languages with a lot of training audio tend to come out best.
Published benchmarks such as FLEURS test the same system on many languages, and the scores usually spread widely from the strongest to the weakest.
Many languages in the list does not mean every language is covered: a language with little training data can be listed and still read poorly.
A multilingual system must either detect the language or be told it, and a wrong choice at that step harms every word after it.
Speech that mixes two languages is harder than speech in one, even for a model that handles both separately.

What it looks like

// illustrative output of a multilingual recogniser
{
  "language": "de",
  "text": "Guten Morgen, wir beginnen mit dem Bericht.",
  "language_confidence": "high"
}
// the same system, given a clip in French, would return language "fr"

See the full list of languages, the tools and the glossary.

Frequently asked questions

Is a multilingual model better than one model per language?

Not always. It is easier to run and covers more languages, but a model built for one language can do better on that language. It depends on the data.

Does it recognise every language?

No. Coverage has limits, and quality differs between languages. Check published test results for your language before trusting a long transcript.

How many languages does the Potok bot handle well?

About 30 languages have good published results, and a few more come with a warning. The language list on this site shows which is which.

Does the bot translate the transcript?

No. The transcript stays in the language that was spoken. The bot does not turn it into another language.

What should I do for a rare language?

Test a short clip first, read the result closely and expect names and technical words to need hand corrections.

Sources

Each source was opened by hand on the date shown. Pages change, so open the source if a detail matters to you.

Whisper speech recognition system (Wikipedia), Wikipedia · checked 2026-10-18
FLEURS speech dataset (Google, Hugging Face), Hugging Face · checked 2026-10-18

Try it on your own recording