Arabic speech to text with Potok runs through a Telegram bot and carries a caution. In published tests of the large-v2 model the word error rate for Arabic is 16, above the level of the recommended languages, and dialect speech scores worse. Those are other people's tests, not our measurements.
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
- Dialect is the main risk: casual talk between relatives can sit far from the Modern Standard Arabic of formal broadcasts.
- A separate third-party average for the turbo model was 10.57, so even the published figures do not settle on one number.
- A word error rate counts words that were swapped, dropped or added compared with a reference text.
- A single speaker in a quiet room is the safest first test.
What a transcript looks like
[00:00] Speaker 1: لنبدأ الاجتماع. أولاً سنراجع نتائج الأسبوع الماضي، ثم نخطط للخطوات التالية.Meaning in English: Let's start the meeting. First, we'll review last week's results, then plan the next steps.
Written by us to show the script and the layout of a transcript. It is not the output of the recognition engine.
Published quality
| Language | Verdict | Published result | Notes |
|---|---|---|---|
| Arabic (العربية) | Use with care | word error rate 16% (large-v2, FLEURS) Whisper paper, TheWhisper benchmark | Published tests show a higher error rate for this language. Check a short sample first and plan to proofread. The figure is from the earlier large-v2 model and a third-party turbo average was 10.57. Dialects differ a lot from Modern Standard Arabic, and error rates are higher on dialect speech. |
These numbers come from published tests by other people, not from tests by Potok. Your own audio, with its noise, accents and microphone, will differ.
How to do it
- Cut a short sample. Take about a minute from the middle of your recording, not the polished intro, and keep the speaker you care about.
- Send it to the bot. Open @potok_now_bot in Telegram and send the clip as a voice message, audio file or video.
- Compare with your ear. Listen while reading. Check names of people and places first, since those are the words a model has seen least often.
- Send the full file. Up to 2 GB is accepted. For a long recording use the Upload button of the bot's mini app.
- Export for reading. Choose DOCX or TXT for review and SRT or VTT for captions. Keep timestamps on so you can jump back to doubtful lines.
| Feature | Details |
|---|---|
| Where it works | Telegram bot @potok_now_bot; large files go through the bot's mini app |
| File size | Up to 2 GB |
| Export | TXT, MD, DOCX, PDF, SRT, VTT |
| Speakers and time | Speaker labels and timestamps can be switched on |
| Languages | About 30 languages with good published results; a few more with warnings |
| Translation | The transcript is not translated into another language |
| Dictation | A separate Potok desktop app lets you dictate text into any program; it recognises Russian speech |
Common mistakes
| Mistake | What happens | Fix |
|---|---|---|
| Judging every Arabic dialect by one published score | The number says little about a regional accent, and errors rise on dialect speech. | Run a sample from the actual speaker before committing hours of audio. |
| Opening the exported text in a left-to-right editor | Digits and punctuation land on the wrong side of the line. | Use a word processor with right-to-left support. |
| Quoting 16 as a result on your files | Readers assume it was measured on your audio. | Say it is a figure from published tests of others. |
| Skipping the speaker check on a group call | Overlapping voices merge into one line. | Switch on speaker labels, which appear as numbers. |
More
See the full list of languages, the tools and the glossary.
Frequently asked questions
Why does this page carry a caution?
Published error rates for Arabic are higher than for the recommended languages and climb further on dialect speech, so read the output carefully.
Does it cope with Egyptian or Gulf speech?
The published tests do not split results by dialect. They only note that dialects differ a lot from the standard form, so try a short sample.
Can I send an Arabic video?
Yes. The bot reads video and audio files, voice messages and YouTube links within the 2 GB limit.
Will diacritic marks be written?
Do not count on it. Everyday Arabic text often leaves short vowels unwritten, so add marks yourself if you need them.
What does the 16 stand for?
It is the word error rate, in percent of words, from published tests of the large-v2 model.
Sources
Each source was opened by hand on the date shown. Pages change, so open the source if a detail matters to you.
- Word error rate (Wikipedia), Wikipedia · checked
- FLEURS speech dataset (Google, Hugging Face), Hugging Face · checked
- Whisper paper, Table 13 (large-v2 on FLEURS) · checked
- TheWhisper benchmark (turbo, average over FLEURS, MLS, Common Voice) · checked
Try it on your own 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