Repolode
Submit
antirez

antirez/whisperbot

A Telegram bot written in C that transcribes voice messages to text using whisper.cpp or qwen-asr.

Open-source alternative to:Otter.aiRevTelegram Premium voice-to-text

View on GitHub
Stars
168
Forks
12
Open issues
1
Last commit
8 months ago

What is it?

Whisper/Qwen3-ASR Bot is a Telegram bot that turns voice messages into text. A user sends a voice message to the bot, and it replies with the transcription. The bot downloads the audio, converts it to WAV with ffmpeg, runs a speech recognition backend as an external process, and streams the result back into the chat by editing the message as text arrives.

It supports two backends: whisper.cpp (via whisper-cli) and qwen-asr, a pure C inference implementation of Qwen3-ASR. The author recommends the Qwen3-ASR backend. The bot is built on botlib and uses a thread-per-request design, with ASR jobs serialized by a mutex so that CPU-heavy transcriptions do not run in parallel. When the queue is short it uses the higher-quality model, and when it grows it switches to a faster model and rejects new requests past a limit.

Backend selection, model paths and limits are set with compile-time defines at the top of whisperbot.c. Pending requests are held in memory only, so they are lost if the bot restarts.

Key features

  • 01Transcribes Telegram voice messages and replies with the text.
  • 02Supports whisper.cpp and qwen-asr as selectable ASR backends.
  • 03Converts incoming audio to WAV with ffmpeg before recognition.
  • 04Streams partial transcription results by editing the reply message.
  • 05Switches to a faster model when the request queue grows.
  • 06Configured through compile-time defines in the C source.

Use cases

  • Running a private voice-to-text bot for a Telegram group on a small server.
  • Experimenting with local speech recognition models behind a chat interface.
  • Using a C-based bot as a reference for botlib-style Telegram bots.

Topics

#telegram-bot#speech-to-text#whisper#asr#voice-transcription

FAQ

What is whisperbot?
A Telegram bot written in C that transcribes voice messages to text using whisper.cpp or qwen-asr. Whisper/Qwen3-ASR Bot is a Telegram bot that turns voice messages into text. A user sends a voice message to the bot, and it replies with the transcription. The bot downloads the audio, converts it to WAV with ffmpeg, runs a speech recognition backend as an external process, and streams the result back into the chat by editing the message as text arrives.
Is whisperbot free to use? What license does it have?
whisperbot has no declared license. Its code is public, but without a license you have no explicit permission to reuse it; ask the author before relying on it.
Is whisperbot still maintained?
No commits for 6+ months. The last commit to whisperbot was on Feb 12, 2026.
What can I use whisperbot for?
Running a private voice-to-text bot for a Telegram group on a small server. Experimenting with local speech recognition models behind a chat interface. Using a C-based bot as a reference for botlib-style Telegram bots.
Where can I get whisperbot?
The source code and installation instructions are on GitHub: https://github.com/antirez/whisperbot

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects