Yts is an open source Elixir Phoenix LiveView prototype that takes a YouTube video and generates a textual summary. It uses modern ML tools like OpenAI Whisper and Facebook's BART for transcription and summarization.
The video used in the preview is 6.27 MB in size and spans 10mins 18s. On my local machine on CPU only it took 1min 14s to download and process. It can be optimized with a faster internet connection and running the ML tasks on the GPU.
When a URL is entered, it is validated and the YouTube video ID is extracted to display the video thumbnail. Meanwhile, the server downloads the video using the yt-dlp CLI tool.
Once the download is complete, ffmpeg encodes the audio and splits it into 60-second chunks so that OpenAI Whisper can generate text transcriptions.
All text chunks are then joined and passed through Facebook's BART model to generate a summary.
To start your Phoenix server:
- Run
mix setupto install and setup dependencies - Start Phoenix endpoint with
mix phx.serveror inside IEx withiex -S mix phx.server
Now you can visit localhost:4000 from your browser.
Ready to run in production? Please check our deployment guides.
- Elixir: https://elixir-lang.org/
- Phoenix/Liveview: https://www.phoenixframework.org/
- yt-dlp: https://github.com/yt-dlp/yt-dlp
- ffmpeg: https://ffmpeg.org/
- Bumblebee: https://github.com/elixir-nx/bumblebee
- Bart: https://huggingface.co/facebook/bart-large-cnn
- Whisper: https://huggingface.co/openai/whisper-large-v3-turbo