This repository contains code of a simple Discord bot that reacts to voice messages, transcribes them, and sends the transcription as a reply to the original voice message.
The speech-to-text model used is a pre-trained OpenAI's Whisper model (specifically large V3), using the code from SYSTRAN/faster-whisper.
- Install Python (3.11 at least)
- Install Poetry
- Install dependencies:
poetry install - Activate the Poetry-created virtualenv:
poetry shell - Set
DISCORD_TOKENenvironment variable to your Discord Bot's Token. - Run the code:
python main.py
The bot reacts to discord messages that have an audio attachment with .ogg extension.
The bot supports the following environment variables:
DISCORD_TOKEN(required) - the token used to authenticate the botMODEL_NAME- the name of the model to be loaded byfaster-whisperfrom Hugging Face Hub. Refer to the original repository to learn more about available pre-trained models. Default:large-v3LANGUAGE- the language for which transcription should be done. If not set, the language is detected for each transcription.
Currently, the model is configured to run on the CPU. For CUDA-enabled deployments, refer to the original faster-whisper repository.