- Summarizes a youtube video provided yt video url and file name.
- Customization options such as word limit, video resolution, bullet points are available.
- Also, provides an option to download the yt video or its audio track.
yt_vid_summarizer_demo.mp4
- Python 3.10 (Might work with higher versions as well)
- FFMPEG:
sudo apt install ffmpeg - Nvidia drivers installed along with CUDA for whisper
- Nvidia GPU with VRAM atleast 6GB (medium) or 10GB (large-v2)
- CPU can also be used but is awfully slow
- Clone the repo and move to the root dir.
git clone https://github.com/prashkrans/yt_video_summarizer.git
cd yt_video_summarizer/
- Create a python virtual environment.
python3 -m venv env_vid_sum
source env_vid_sum/bin/activate
- Install the requirements (Might take some time).
pip install -r requirements.text
- Add the GROQ_API_KEY line to ~/.bashrc and then source it.
nano ~/.bashrc
export GROQ_API_KEY='gsk_xxxxxxxxxxxxxx'
source ~/.bashrc
Note: GROQ_API_KEY could be replaced with dot_env or direct api key in line 9 of _4_summarize.py
4. Run the gradio app and Ctrl+LeftClick on the local host link to use the gradio app in your browser.
python3 main_gradio.py
- When the gradio app has started, provide the YT video url and file name.
- Select a language in which the summary is required.
- Select the word limit.
- Select "Summarize as bullet points" if the summary is to be created as bullet points.
- Select the video resolution i.e. SD, HD or FHD
- Click on "Summarize Video" and wait for processing to end.
- Upon completion, two different summaries and the original transcription would be presented with the option to download them.
- It uses Whisper medium/large-v2 to generate transcription.
- It uses mixtral and llama3.1_70b (hosted on Groq) LLMs to summarize the transcription generated.
This gradio app and Whisper's code and model weights are released under the MIT License. See LICENSE for further details.