# Music data and model attribution

## Catalog

Metadata, artwork and preview URLs: Apple Music / iTunes public catalog. Provider terms apply. No commercial song recordings are bundled. Preview clips are fetched from their original provider; audio plays only on request.

## MetaBrainz context

Sources: MetaBrainz Foundation and contributors — [MusicBrainz](https://musicbrainz.org/), [ListenBrainz](https://listenbrainz.org/) and [AcousticBrainz](https://acousticbrainz.org/). Retrieved November 2026. MusicBrainz core identifiers are CC0. MusicBrainz supplementary tags and their normalized derivative fields retain [CC BY-NC-SA 3.0](https://creativecommons.org/licenses/by-nc-sa/3.0/); this does not license other providers' data. See the [MusicBrainz data license](https://musicbrainz.org/doc/About/Data_License).

Changes: matched recording and artist identities; normalized and filtered style tags and listening relationships; quantized existing full-recording descriptors.

## Song-specific styles

Factual musical-style metadata from [DBpedia](https://www.dbpedia.org/resources/knowledge-graphs/) and its Wikipedia sources. DBpedia/Wikipedia attribution and share-alike terms apply to their source content and derived style fields. Artist and title matching is conservative; album/artist context is distinct from song-level sound analysis.

## Music models

Unmodified pretrained networks from the Music Technology Group, Universitat Pompeu Fabra:

- [Discogs EffNet](https://essentia.upf.edu/models/feature-extractors/discogs-effnet/) — Alonso-Jiménez, Serra and Bogdanov, “Music Representation Learning Based on Editorial Metadata from Discogs,” ISMIR 2022.
- [MTG Jamendo mood/theme](https://essentia.upf.edu/models/classification-heads/mtg_jamendo_moodtheme/).
- [MTG Jamendo instruments](https://essentia.upf.edu/models/classification-heads/mtg_jamendo_instrument/).

These distributed model files are unchanged. See [Essentia's model licensing](https://essentia.upf.edu/models/) for CC BY-NC-ND 4.0 and commercial licensing information. Song Atlas pools model outputs over the entire available preview, then projects and normalizes the numerical descriptors. Mood/style estimates are not lyrical interpretations.

## Browser runtime

[ONNX Runtime Web](https://github.com/microsoft/onnxruntime), Microsoft Corporation and contributors, version 1.23.2. MIT; see [included license](vendor/LICENSE-ONNX-Runtime.txt).