Telegram ASR Bot

This repository contains a Python script for a Telegram bot that uses NVIDIA's NeMo ASR models to transcribe voice messages. The bot can handle both locally stored models and pre-trained Hugging Face models.

Overview

• Purpose: Transcribe voice messages sent to the Telegram bot.
• Key Features:
  • Supports local NeMo ASR models and Hugging Face models.
  • Converts .ogg voice messages to .wav format for transcription.
  • Deletes temporary .ogg files after processing.

Configuration

The bot requires a configuration file in the config/ directory. Use config.yaml as the default configuration file. Below is an example configuration:

config/config.yaml

# Path to the local model file. This will be used first if it exists.
model_path: '/path/to/alternate/model/if/needed.nemo'

# Name of the Hugging Face model to fall back on if the local model is not available.
hf_model_name: 'nvidia/stt_hy_fastconformer_hybrid_large_pc'

# Token for the Telegram bot to interact with your work PC.
telegram_token: 'your_bot_token'

Notes:

• Replace /path/to/alternate/model/if/needed.nemo with the actual path to your local NeMo ASR model.
• Replace your_bot_token with the Telegram bot token.
• If you want to use a different configuration file (e.g., for another environment), create it in the config/ directory and update the config_path in the script accordingly.

How to Run the Bot

Running with run.sh

1. Use the run.sh script to automatically set up the Docker environment, build the Docker image, and start the container:

   ./run.sh

   This script will:

   • Build the Docker image.
   • Stop and remove any existing containers with the same name.
   • Start a new container for the bot.

2. Monitor the logs to ensure the bot is running:

   docker logs -f armenian-asr-bot

3. Interact with the bot via Telegram:

   • Send /start to the bot.
   • Send a voice message, and the bot will transcribe it.

Functionality

• Starting the Bot:
  • Send /start to the bot to begin.
• Sending a Voice Message:
  • Send a voice message, and the bot will:
    1. Download the .ogg file.
    2. Convert it to .wav.
    3. Transcribe the message using the NeMo ASR model.
    4. Respond with the transcribed text.

Workspace

The workspace/ directory is used to store temporary user files (e.g., .ogg and .wav files). Subdirectories are created for each user based on their Telegram ID.

Cleaning Up

The bot automatically deletes .ogg files after conversion to .wav.

Support

For issues or questions, please contact the repository maintainer.