88,732 tracks · 114 genres · one shared audio-feature space
The Sonic Genome — Genre Galaxy. UMAP projection of n = 88,732 Spotify tracks (2022 snapshot) from 9 standardized audio features (danceability, energy, valence, acousticness, instrumentalness, liveness, speechiness, loudness, tempo). Points are colored by super-genre family and sized by popularity (0–100); diamonds mark the median projected position of each of the 114 genre seeds, with up to 25 of the most-tagged seeds labeled (greedy spacing filter for legibility). Draw order stacks larger families beneath smaller ones so rare families stay visible. Acoustic/instrumental territory separates sharply from the electronic and metal mainland, falsifying the no-signal hypothesis of a single interleaved cloud. Caveat: UMAP axes are non-metric — distances and directions are not interpretable as feature differences, and cluster sizes are not meaningful. Toggle families via the legend; drag to pan, scroll to zoom; hover for track, artist, and full genre tags.
Projection: umap-learn 0.5.12, n_neighbors = 30, min_dist = 0.1, metric = euclidean on z-scored features, fixed seed random_state = 20260610 (never auto-tuned). 1009 tracks excluded by quality flag (tempo/time-signature detector failures). Lineage: data/processed/track_embedding_2d.lineage.json · Source: Kaggle maharshipandya/-spotify-tracks-dataset (Spotify deprecated the /audio-features endpoint in Nov 2024; this 2022 snapshot is canonical).