discord bot for transcribing voice-chat conversations and then summarizing them using language models. Supports both running models locally (phi-2 and llama 3.2 1B) and using cloud hosted apis (OpenAI gpt-4o-nano)
The summarization has partially been made to quite primitive and dumb as it is usually funnier than way. If you want proper, professional summarizes edit the example_summaries.txt file to suit your preferences and use GPT4 (and maybe something smarter than the nano model).
creating a python virtual environment is highly recommended. I personally had issues with crackling/hacky audio before doing so, so there seems to be a risk of insidious dependency conflicts.
python -m venv venv
Windows:
.\venv\Scripts\Activate.ps1 (powershell, if using normal command prompt run activate.bat instead)
Linux/MacOS:
source venv/bin/activate
then the virtual environment will be set up and you can install the requirements and run with the following three commands:
pip install torch==2.2.2+cu118 torchvision==0.17.2+cu118 torchaudio==2.2.2 --extra-index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
py bot.py
see config.py for configuration and make sure to fill up .env with secrets.