whisper_tg_bot

Telegram bot and user bot that utilizes the Whisper model for transcription of voice messages and video notes. It is designed to run with arguments such as token, CPU threads, and model size. The bot uses Pyrogram library for interaction with Telegram, and it employs asyncio and threading for asynchronous and parallel processing.

Initial Setup

1. Clone the repository: Clone this repository using git clone.
2. Create Virtual Env: Create a Python Virtual Env venv to download the required dependencies and libraries.
3. Download Dependencies: Download the required dependencies into the Virtual Env venv using pip.

git clone https://github.com/ergolyam/whisper_tg_bot.git
cd whisper_tg_bot
python -m venv .venv
.venv/bin/python -m pip install uv
.venv/bin/python -m uv sync

Deploy

• Run the bot:

  TG_TOKEN="your_telegram_bot_token" CPU_THREADS="4" MODEL_SIZE="tiny" uv run main.py
  

• Other working env's:

  LOG_LEVEL="INFO"
  MODE="bot" #bot, user, mixed
  TG_ID="your_telegram_api_id"
  TG_HASH="your_telegram_api_hash"
  TG_TOKEN="your_telegram_bot_token"
  CPU_THREADS="number"
  MODEL_SIZE="tiny"
  #tiny, tiny.en
  #base, base.en
  #small, small.en
  #medium, medium.en
  #large-v1, large-v2
  #large-v3, or large
  

Container

• Pull container:

  podman pull ghcr.io/ergolyam/whisper_tg_bot:latest
  

• Deploy in container as bot:

  mkdir -p $HOME/whisper_cache/ && \
  podman run --tmpfs /tmp \
  --name whisper_tg_bot \
  -v $HOME/whisper_cache/:/root/.cache/huggingface/:z \
  -e MODE="bot" \
  -e TG_TOKEN="your_telegram_bot_token" \
  -e MODEL_SIZE="tiny" \
  -e CPU_THREADS="4" \
  ghcr.io/ergolyam/whisper_tg_bot:latest
  

• Deploy in container as user bot:

  mkdir -p $HOME/whisper_cache/ && \
  mkdir -p $HOME/sessions/ && \
  podman run --tmpfs /tmp \
  --name whisper_tg_bot \
  -v $HOME/whisper_cache/:/root/.cache/huggingface/:z \
  -v $HOME/sessions/:/app/sessions/:z \
  -e MODE="user" \
  -e TG_ID="your_telegram_api_id" \
  -e TG_HASH="your_telegram_api_hash" \
  -e MODEL_SIZE="tiny" \
  -e CPU_THREADS="4" \
  ghcr.io/ergolyam/whisper_tg_bot:latest
  

use MODE="mixed" to combine modes

Features

1. Transcribes voice messages and video notes into text using the Whisper model.
2. Supports multiple Whisper model sizes for transcription.
3. Utilizes multithreading for efficient processing of voice messages and video notes.
4. Selection of the bot operating mode, as a user bot or as a regular bot, as well as mixed mode.