Romanian speech to text in Potok is a recommended language, and published tests of the large-v3 model give a word error rate of 8.2. The figure comes from other people's work, not our measurement. Send voice notes, audio or video to the bot, then read the result.
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
- Two separate published sources back the figure, which makes it a steadier data point than a single test.
- Romanian uses five marked letters, and some systems write them with cedillas instead of commas, so a search in your text may miss words.
- Definite articles attach to the end of the noun, so word endings carry meaning worth proofreading.
- At 8.2 expect an error in roughly one word out of twelve against a human reference.
What a transcript looks like
[00:00] Speaker 1: Să începem ședința. Mai întâi vom analiza rezultatele săptămânii trecute, apoi vom planifica următorii pași.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 |
|---|---|---|---|
| Romanian (Română) | Recommended | word error rate 8.2% (large-v3, FLEURS) Canary-1B-v2 paper, Whale paper | Published tests show a low error rate for this language. |
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
- Prepare the file. Voice notes, calls and lectures all work. Keep the original rather than a re-encoded copy, and note the speakers' names beforehand.
- Hand it to the bot. Open @potok_now_bot and send the file or a link.
- Use Upload for the heavy ones. Up to 2 GB is allowed through the Upload button.
- Standardise diacritics. Run a find-and-replace so every marked letter uses one form.
- Read endings. Scan articles and plural endings on nouns, where small changes alter meaning.
- Choose a format. TXT, MD, DOCX, PDF, SRT or VTT.
| 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 |
|---|---|---|
| Searching a transcript without checking diacritic forms | Two encodings of one letter hide matches. | Normalise the letters once, then search. |
| Reading 8.2 as exact | It is a published figure for other audio. | Check names, numbers and quotes. |
| Sending a recording with music under the voice | Music masks consonants. | Use a version without the music if one exists. |
| Skipping speaker labels in an interview | Questions and answers merge. | Turn labels on first. |
More
See the full list of languages, the tools and the glossary.
Frequently asked questions
What is the published score?
A word error rate of 8.2 for the large-v3 model, reported by others.
How many sources give the number?
Two published sources were used for Romanian, which makes the figure steadier than one based on a single test.
Why do some letters look different?
Some systems write the marked letters with cedillas and others with commas, so normalise them.
Does the bot change the language of the transcript?
No. The text stays in Romanian, in the language of your recording, with no second language added.
Can I get captions?
Yes. Turn timestamps on and export SRT or VTT.
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
- Canary-1B-v2 paper, Table 11 (Whisper large-v3 on FLEURS) · checked
- Whale paper, Table 5 (Whisper large-v3 on FLEURS) · 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