CoolFace
Datasetpublic

Ketose333/music-mood-recs-assets

music-mood-recs — derived assets 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. Source & license 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.

sourceHugging Faceotherupdated 3mo agoView on Hugging Face
0likes34downloads
Dataset Card

music-mood-recs — derived assets

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.

Source & license

All audio originates from the MTG-Jamendo Dataset (mood/theme subset, 50-TAR sample, 3,585 tracks): https://github.com/MTG/mtg-jamendo-dataset

Per the dataset's own terms:

  • —Metadata: CC BY-NC-SA 4.0
  • —Audio: each track carries an individual Creative Commons license chosen by its Jamendo uploader. See audio_licenses.txt for the authoritative per-track artist credit + license.
  • —The MTG-Jamendo Dataset is made available solely for non-commercial research and academic use. Any other use, including commercial use, requires prior written authorization from Jamendo S.A. (contact hello@jamendo.com).

This mirror is not an independent redistribution channel for these tracks — it exists solely to support the non-commercial coursework use case above, within the same scope the original dataset permits. Refer to audio_licenses.txt (linked above) for per-track artist attribution.

Contents

  • —data/audio/ — audio-low MP3s for the 3,585-track subset
  • —artifacts/melspecs/ — precomputed log-mel spectrograms (.npy)
  • —artifacts/embeddings.npy — CNN classification-embedding matrix (64-dim per track)
  • —artifacts/melspec_manifest.csv, artifacts/subset_meta.csv — track manifests

Note: this mirror only hosts the bulky derived data the Streamlit app needs at runtime. Report figures and the project's own download/training scripts live in the main repo (Ketose333/music-mood-recs), not here.