Skip to content
HEHunter Eppley
All demos
Music · Audio features

The Sonic Genome

Pythonpandasscikit-learnUMAPPlotly

Every track carries a set of measurable audio features: energy, valence, danceability, acousticness, tempo, and more. On their own those numbers are hard to read. The Sonic Genome treats the whole catalog as data and turns it into something you can actually navigate.

I reduced the high-dimensional feature space down to two dimensions with UMAP so similar-sounding music sits close together, then mapped the same data four ways: the overall landscape, the mood split, the signature of each genre, and how the features move together. Each figure below is interactive. Hover, zoom, and pan.

01Genre GalaxyFull screen

Every track placed by sound. The high-dimensional feature space is projected to two dimensions with UMAP, so neighbors here are neighbors in sound.

02Mood QuadrantFull screen

Valence (sad to happy) against energy, sorting tracks into mood quadrants from calm and melancholy to bright and intense.

03Genre FingerprintsFull screen

The average audio-feature profile of each genre, drawn as a fingerprint, so you can see what actually makes one genre sound different from another.

04Feature CorrelationFull screen

How the audio features move together, as a Pearson correlation matrix. Loud tracks tend to be energetic; acoustic tracks tend not to be.

Dataset

Built from per-track audio-feature data across many genres.

Want a dataset turned into something like this? Get in touch