This is a simple demo that visualizes songs in audio embedding space using MERT (Music Understanding Model). It demonstrates how AI models can learn representations of music that capture acoustic and stylistic similarities.
The demo analyzes 11 songs spanning different eras and genres:
- US Innovators (1950s): Chuck Berry, Elvis Presley
- British Invasion (1960s-70s): The Beatles, The Kinks, Rolling Stones
- 80s Pop: Madonna, Prince
- Modern Pop (2010s-2020s): Billie Eilish, Tate McRae, The Weeknd
The visualization uses t-SNE and UMAP to project high-dimensional MERT embeddings into 2D space.
Songs are already in the data/songs/ folder. To generate the visualizations:
uv run python audio/musical_diffusion_demo.pyThis will create two PNG files in the figures/ folder showing songs in 2D embedding space (t-SNE and UMAP projections).
See audio/README.md for more details.