Transformer model learns to imitate jazz pianists style accurately
Learning Jazz Pianist Style with Cross-Attention Conditioning
SoundMachine Learning
Summary
Jazz pianists have unique ways of playing that experts can recognize quickly, but it’s hard to describe exactly what makes their style special. The authors studied how a type of AI called a transformer can learn and identify these styles from music data. They enhanced the AI to generate new jazz piano music that sounds like a specific pianist by giving it clues about that artist’s style. Tests showed the AI can produce music that listeners and machine classifiers correctly match to the right pianist, even finding the unique moments that define each artist’s sound.
What this means in practice
- •For music software developers: Create music generation tools that produce jazz piano pieces in the style of specific famous pianists.$Commercial implications: This enables selling personalized AI-based jazz composition tools or plugins for musicians wanting stylistic authenticity.
- •For musicologists and archivists: Automatically identify and analyze stylistic signatures in jazz piano recordings to assist in music study and archiving.
Authors
Drew Edwards, Akira Maezawa, Simon Dixon
Abstract
Jazz pianists develop distinctive traits that experienced listeners can often identify within seconds, yet the features underlying this recognition resist formal description. We study jazz pianist style through the lens of a pretrained symbolic music transformer, showing that its learned representations already encode pianist identity well enough for highly accurate classification across two benchmarks. We then augment the transformer with cross-attention over learned pianist identity embeddings, enabling it to generate music conditioned on a specific artist's style. Two evaluation protocols confirm that the generator captures meaningful stylistic structure: a sliding-window classifier consistently attributes conditioned continuations to the correct artist, far above unconditioned baselines; and a classifier trained entirely on synthetic generations identifies real pianists across 12 classes with 87% chunk-level and 95% song-level accuracy. Finally, we repurpose the classifier to locate the most characteristic moments within a performance, surfacing the specific musical gestures that distinguish each pianist's voice.