Audio, mel-spectrograms, and embeddings for a non-commercial student deep-learning
coursework project (audio mood classification + content-based recommendation, single
CNN model). This repo exists only so the project's Streamlit demo app can fetch these
files at runtime without bundling ~7GB into the git repo.
All audio originates from the MTG-Jamendo Dataset (mood/theme subset, 50-TAR sample,
3,585 tracks):… See the full description on the dataset page:
https://huggingface.co/datasets/Ketose333/music-mood-recs-assets.
