How to turn a long voice message into text

Potok transcribes Russian speech; this site has been translated into English for convenience.

Telegram's built-in button turns a long voice message into text, but without Premium it works only a few times a week. For voice messages of any length with no such limit there is the Potok Telegram bot @potok_now_bot: send the recording, a short one comes back as text right away, a long one as a file. The free allowance is 60 minutes of recognition per month.

Transcribing a long voice message by hand, listening, stopping and retyping, takes far longer than simply forwarding the recording to a bot. The Potok Telegram bot @potok_now_bot handles it in one go, however long the voice message is.

How much effort a manual transcription takes

Transcribing a long voice message by hand is not just typing what you hear. You keep stopping the recording, rewinding to a word you did not catch, and listening again for the intonation if the speech is fast or halting. Punctuation and capital letters are added by hand, from memory of the pauses in the voice. The longer the recording, the more such stops there are, so the job stretches over several sittings instead of one continuous listen. Automatic transcription does all of this in one pass: the text arrives in full with the punctuation already in place, and you only need to listen to the recording itself to check the doubtful spots.

How to turn a voice message into text faster

1. Save the voice message as a file if it did not arrive in Telegram, so that you can forward it to the bot.
2. Open the Telegram bot @potok_now_bot and press "Start" (the /start command) if you are writing to it for the first time.
3. Forward the voice message to the bot directly in a private chat, or send the saved file.
4. The bot recognizes the whole recording in one go, regardless of its length.
5. A short transcript arrives as text right in the chat, a longer one as a separate file. You can choose the format, TXT or DOCX, beforehand in the settings.
6. Timecodes and speaker labeling, "Speaker 1" and "Speaker 2", are also turned on in the /settings.

What to check in the finished text

• the bot adds punctuation and capital letters itself when it proofreads, but it is worth skimming the text, since the model can make a mistake on a rare word or name;
• if the recording has several voices and speaker labeling was not turned on beforehand, the lines will not be separated, and you will have to relisten and rearrange them by hand;
• the bot does not shorten or retell the text, so the transcript contains as many extra words and repetitions from the speech itself as the recording does;
• if you need subtitles or a file for editing, the "Other format" button gives the same text as SRT and VTT, which are easy to check against the recording by timecodes.

A more detailed comparison of transcription services is on the page audio-to-text transcription: a comparison of services.

When the bot will not fit

If the voice message is not in Russian, the Potok bot will not transcribe it: there is only one recognition language, and there will be no translation. It will not fill in the real names of the speakers either; in a text with labeling they appear as "Speaker 1", "Speaker 2". Transcription is only possible for a finished recording: the bot cannot listen to a conversation in real time and turn it into text at once. It also does not accept links to Instagram and TikTok, and among video platforms only YouTube, VK Video and RuTube. The difference between this kind of transcription and ordinary voice text input is covered on the page transcription and voice input: what is the difference. The bot is a cloud service: it does not work without an internet connection, and processing runs on servers in Russia.

Frequently asked questions

Primary sources: core.telegram.org · www.ixbt.com · developers.sber.ru · t.me