Computer Models Identify Jazz Legends’ Musical Imprints

Every supremely talented musician leaves behind a unique, personal signature in their performance, but many of these nuanced interpretation patterns remain too minute for even trained human ears to catch. To decode these artistic identities, researchers from the University of Cambridge in Cambridge, England, trained machine learning models to identify the unique “sonic fingerprints” of jazz legends.

The study analysed 84 hours of audio, spanning 1629 performances by 20 iconic jazz pianists, including Bill Evans, Chick Corea and Thelonious Monk. To enable computational analysis, the research team converted the recordings into digital files, providing a detailed visual and mathematical map of pitch, timing, melody, harmony, rhythm and dynamics.

The best-performing AI model identified the correct pianist with an accuracy of 94.4%. This included successfully pinpointing specific stylistic signatures, such as Bill Evans frequently using a descending major or minor seventh arpeggio, or Oscar Peterson favouring rapid octave tremolos.

By processing vast datasets with statistical rigour beyond manual human capabilities, this approach opens new doors for arts scholarship. The technology can help identify unattributed works, detect art forgeries, and assist music students in analysing master techniques.

The researchers have also released a free web application, allowing jazz fans to visually explore and listen to the distinct stylistic signatures of these icons.