When learning from watching videos, I often find it time consuming to navigate through a long video to find the part that I am interested and learn from it quickly. This project allows you to download a YouTube video, extract its audio transcription, and capture representative frames from the video. The extracted information and frames are then compiled into an HTML report for easy viewing.
- Downloads YouTube videos in the best available quality.
- Transcribes audio to text using OpenAI's Whisper model.
- Extracts representative frames from the video at specified intervals.
- Generates a neatly formatted HTML report with sections of text and corresponding frames.
This code is tested in a MacOS Sequoia (version 15.1) environment with following packages
- Python 3.12 or higher
yt-dlpwhisperopencv-pythonnumpyjinja2
-
Clone this repository:
git clone https://github.com/yourusername/youtube-video-processor.git cd youtube-video-processor -
Install the required packages:
pip install -r requirements.txt
note that for it is better to install using
pip install git+https://github.com/openai/whisper.git; see openai/whisper#251
To run the video processor, use the following command:
python main.py --youtube_url "https://www.youtube.com/watch?v=Mn_9W1nCFLo"Replace the URL with the desired YouTube video link.
--youtube_url: The URL of the YouTube video you want to process.
The output will be saved in the output directory as vid2doc.html. This file will contain:
- The title of the video.
- Sections of transcribed text with links to the corresponding time in the video.
- Extracted frames from the video.
Pleas reference
This project is licensed under the MIT License. See the LICENSE file for details.