Transcribing audio to text with a Telegram bot and in the browser

Potok turns speech into text in 33 languages, including English and Russian.

The easiest way to transcribe audio to text is Potok's Telegram bot @potok_now_bot: forward a voice message or send a file, and the bot returns the text with punctuation. For large files, the bot's mini app does the same transcription in the browser. 60 minutes of recordings per month are available for free; for typing your own text by voice on a computer there is a separate Potok app.

A finished recording, such as an interview, a lecture or a conversation from a voice recorder, can be turned into text without installing programs: the file or link is sent to a messenger or uploaded to a website, and the text comes back in a minute or two.

Transcribing audio to text with a Telegram bot and in the browser

The whole process comes down to sending the recording and waiting for the reply.

1. Open Potok's Telegram bot @potok_now_bot and press Start (the /start command).
2. Forward a voice or audio message from any chat, or send an audio file straight to the bot, including as a document.
3. Wait for the reply: a short recording comes back as text in the chat, a long one as a file; the file format (TXT or DOCX) is chosen in the settings.
4. Under the text, the buttons Summary, Tasks and deadlines, Ask and Other format appear; choose the kind of result you need.
5. If you need timestamps or speaker labels, turn them on in advance in the bot settings (/settings).
6. For a larger file, open the bot's mini app, sign in by email or via Telegram and drag the file into the window; you can also paste a YouTube, VK Video, RuTube, Instagram or TikTok link there; if the connection drops, the upload continues from where it stopped.
7. Finished transcripts open in the My recordings section at any time: access is shared between the bot and the web version, and a recording started in Telegram is also visible in the browser.

How to check that the bot transcribed everything correctly

Before delivering a long recording, the bot proofreads it itself: it places punctuation and capital letters and removes obvious recognition errors. The text is not shortened or retold: the transcript contains all of the audio, not a summary. The easiest way to check a specific place against the original is in the web version: it has a player, and clicking a phrase plays the recording from exactly that point. If it is easier to get a direct answer than to read the whole text, there is an Ask button under the transcript that answers a question about the recording's content. All transcripts are stored in the My recordings section and can be opened again at any time, so you can check them not right away but when you have time. With a video recording the check works the same way; a detailed guide is in the article how to turn a video into text.

When it is better not to send a recording to the bot

Potok's bot is a cloud service: recognition runs on servers in Russia, and without an internet connection it does not work. A recording in another language works too, the bot understands 100+ languages (the language is detected automatically), but it does not translate. For non-Russian speech there are separate services, for example Otter.ai, which supports English, Spanish, French, German, Japanese and Chinese, but not Russian. The bot tells the participants of a conversation apart only by number, Speaker 1, Speaker 2, and does not know their names: for minutes with real names you will have to type them in by hand after transcription. Transcription during a call is not its scenario either: the recording is first saved as a file and only then forwarded to the bot. If choosing a tool matters more for your task than dealing with one recording, the comparison of transcription services is useful.

Frequently asked questions

Primary sources: speech2text.ru · audio-2-text.ru · turboscribe.ai · turboscribe.ai