Abstractive Text Summarization from Videos

An abstractive summarization web application that uses Google T5-small fine-tuned model to generate summaries from videos. You can simply paste the YouTube video link into the input section or upload an mp4 file.

Why?

• Third-part summarization tools is pain in the a**
• Almost all tools are pretty expensive and the ones that are available are not very great.
• Privacy concerns and data sharing (ex: ChatGPT)
• I couldn't find a simplified solution and had an engineering mindset. So made one myself. Would be happy if you use it or contribute to it :)

🚀 Quickstart

Start by cloning this repository

git clone https://github.com/siddheshtv/abstractive-summarization.git

Next, cd into the backend directory and install the required libraries

pip install -r requirements.txt

Run the following command to start the backend server that uses FastAPI and uvicorn

uvicorn main:app --reload

Hint: To check if the backend server is up and running, you can send a GET request to http://0.0.0.0:8000/. This will tell you if the server is running, and if not, it will throw an error. 4. Next, cd into the frontend directory and run:

npm install

Finally, start the react frontend using:

npm start

🤝 Contributing

Clone the repo

git clone https://github.com/siddheshtv/abstractive-summarization.git

Install the requirements

$backend> pip install -r requirements.txt
$frontend> npm install

Issues to solve at this point

• Developing a better transcription model
• Developing a better "abstractive" summarization model
• Enhance the user interface
• Using custom user authentication (instead of current: Auth0)
• Anything else that might seem interesting to you (incl. code optimization, handling concurrent requests, parallization, etc.)

Submit a pull request

If you'd like to contribute, please fork the repository and open a pull request to the main branch.