Hindi transcription: audio and video to text

By the Potok teamUpdated ~2 min read

Potok can turn Hindi audio and video into text in Telegram, with a caution because published numbers disagree. The model announcement gives a word error rate of 16.9, while independent tests with different text normalisation reported about 37. Files up to 2 GB are accepted.

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

  • The announcement figure of 16.9 is a word error rate, and it is a published number, not our own test.
  • Independent tests with a different text normalisation reported about 37, and a third-party turbo average was 19.25.
  • Expect more mistakes than in the languages marked as recommended.
  • Hindi is written in Devanagari, and speech often mixes in English words.

What a transcript looks like

Sample sentence in Hindi
[00:00] Speaker 1: चलिए बैठक शुरू करते हैं। पहले हम पिछले हफ़्ते के नतीजों की समीक्षा करेंगे, फिर अगले कदमों की योजना बनाएँगे।

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

Quality for this language
LanguageVerdictPublished resultNotes
Hindi (हिन्दी)Use with careword error rate 16.9% (large-v3, FLEURS)
Whisper discussion, TheWhisper benchmark
Published tests show a higher error rate for this language. Check a short sample first and plan to proofread.
The figure of 16.9 comes from the model's announcement; independent tests with a different text normalisation reported about 37, and a third-party turbo average was 19.25. Expect more mistakes than in the languages marked as recommended.

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

  1. Start with a minute of audio. Send it to @potok_now_bot before committing a long file.
  2. Mark what you had to change. Count how many lines need edits. That tells you whether 16.9 or something closer to 37 fits your speaker.
  3. Send the full recording. Files up to 2 GB are accepted; use the Upload button of the mini app for long ones.
  4. Look at English words. Check how mixed-in English terms and names were written.
  5. Export. TXT, MD, DOCX, PDF, SRT or VTT. Keep timestamps on for review.
What Potok can do
FeatureDetails
Where it worksTelegram bot @potok_now_bot; large files go through the bot's mini app
File sizeUp to 2 GB
ExportTXT, MD, DOCX, PDF, SRT, VTT
Speakers and timeSpeaker labels and timestamps can be switched on
LanguagesAbout 30 languages with good published results; a few more with warnings
TranslationThe transcript is not translated into another language
DictationA separate Potok desktop app lets you dictate text into any program; it recognises Russian speech

Common mistakes

Common mistakes and how to fix them
MistakeWhat happensFix
Quoting only the best figureYou promise readers fewer errors than the evidence supports.Mention that independent tests reported about 37.
Assuming the text will be in Roman lettersThe result is written in Devanagari and search terms may not match.Search with Devanagari terms.
Skipping review of English insertionsTerms spoken in English may be written in an unexpected form.Compare them with the audio.
Using the transcript unread for official workErrors in a caution language reach readers.Have a Hindi speaker proofread.

More

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

Frequently asked questions

Why does Hindi have a caution?

Published results disagree, from 16.9 in the announcement to about 37 in independent tests, so the real error rate is uncertain.

Why do the numbers differ so much?

The independent tests used a different text normalisation, which changes how mismatches are counted.

Does it handle mixed Hindi and English?

The published figures do not cover mixed speech, so test a clip and check English terms.

Can I export subtitles?

Yes. Export SRT or VTT and proofread them before use.

Is the Hindi translated into English?

No. The text stays in Hindi, and the bot does not translate the transcript into another language.

Sources

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

  1. FLEURS speech dataset (Google, Hugging Face), Hugging Face · checked
  2. Word error rate (Wikipedia), Wikipedia · checked
  3. OpenAI Whisper discussion on large-v3 · checked
  4. 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