About TgVoiceBrief
A Telegram bot that extracts key points from long voice messages.
Why we exist
Voice messages in Telegram have become normal — but listening to 5–10 minute recordings every time isn't feasible. Existing solutions offered transcription: the full text of the voice note. That's better than listening, but still requires reading a wall of 500–1000 words.
We decided to do something different: not transcribe, but understand. Not text, but meaning. TgVoiceBrief uses AI to extract 3–5 key bullet points — the ones the recipient actually needs.
Our mission
To give people back the time they spend listening to voice messages. By our estimates, an active Telegram user listens to 30–60 minutes of voice notes a day. Most of that time is wasted: filler words, repetitions, tangents. We remove them automatically.
Technology
The bot works in two steps: first it recognizes speech using a Whisper-class neural network model, then it passes the text to a language model for meaning analysis and key point extraction. The result is not a shortened transcript, but a semantic summary: what matters, in priority order, no filler.
English and Russian are supported. Processing takes 3–15 seconds depending on voice note length and queue load.
Privacy
Voice messages are processed in real time and are not stored on our servers after processing. We do not share data with third parties or use voice messages for model training.