SummaryTube is a project designed to download YouTube videos, extract text using whisper.cpp (which requires less VRAM than importing Whisper in Python and supports Apple Metal), and then utilize the OpenAI API to summarize the entire video and generate bulleted points.
This project was developed during my free time and is currently tailored to run on macOS. Keep in mind that the codebase might not be optimal, as it was a personal project, and I plan to update it in the future. The future updates aim to handle all extraction and summarization tasks using local models for enhanced efficiency.
It currently only does the text extraction locally, because I have a machine that is not capable Im using online services to summarize, you can use any other LLM to make the software fully local.
SummaryTube leverages the OpenAI API for its multilingual capabilities, a feature often lacking in many open-source language models. Additionally, the OpenAI API proves to be cost-effective, making it a practical choice for this project. It's important to note that this implementation is optimized for small YouTube videos, as extensive testing with long texts and the OpenAI API is pending.
-
Clone the Repository:
git clone https://github.com/hardcod3dd/summarytube.git
-
Navigate to the Project Directory:
cd summarytube -
Install Dependencies: Ensure you have the necessary dependencies installed. You may need to install additional packages for
whisper.cppand other project components.ffmpeg requirements.txt -
Run the Application:
bash start.sh
Future updates to this project will focus on improving the codebase, optimizing the extraction and summarization process using local models, and expanding compatibility.