raw/README.md

Raw dataset metadata

The raw audio is not redistributed in this repository. These files are the upstream metadata and licence tables needed to rebuild the caches faithfully and to know which terms apply to each recording.

Song Describer Dataset

Record: https://zenodo.org/records/10072001

Included here:

• song_describer.csv — captions and track metadata.
• audio_metadata.tsv — Jamendo artist/album/track metadata.
• audio_licenses.txt — the per-track Creative Commons licence and attribution line for each recording.

The dataset's captions are CC BY-SA 4.0. The recordings are not all CC BY-SA: the licence table includes CC BY-NC-ND and CC BY-NC-SA, among others. Check audio_licenses.txt before reusing a recording.

To rebuild the cache, download the audio into data/raw/song-describer/audio/ using the layout described by song_describer.csv, then run:

staleynet prepare  --song-describer data/raw/song-describer --cache data/cache
staleynet tokenize --cache data/cache

JamendoLyrics MultiLang

Repository: https://github.com/f90/jamendolyrics

The upstream README says the dataset has moved to https://huggingface.co/datasets/jamendolyrics/jamendolyrics.

Included here:

• JamendoLyrics.csv — per-song metadata and the LicenseType column.
• LICENSE — the upstream MIT licence for the code and metadata.
• README.md — the upstream dataset description.

The songs and lyrics are not MIT: each carries the Creative Commons licence listed in JamendoLyrics.csv. The raw mp3/ and lyrics/ directories are not included here.

To rebuild the cache after cloning the upstream repository into data/raw/jamendolyrics/:

staleynet prepare  --jamendo-lyrics data/raw/jamendolyrics --cache data/cache-jamendolyrics
staleynet tokenize --cache data/cache-jamendolyrics --reuse data/cache/tokenizer.json

Own recordings

recordings/, overfit/, and cache-recordings/ are the author's own vocal takes. They are included for reproducibility only and are not licensed for reuse. All rights are reserved.