Can computers learn what makes the most iconic jazz musicians stand out?
Machine learning models can identify jazz pianists from recordings and reveal the musical "fingerprints" that make individual performers recognizable, according to a study investigating 20 famous jazz pianists. The research, published in Nature Machine Intelligence, could offer insight into artist attribution and style, cultural heritage, and music education.
Researchers have developed machine learning models capable of recognizing jazz pianists with remarkable accuracy, shedding light on the unique "fingerprints" that distinguish individual musicians. The study, published in Nature Machine Intelligence, analyzed recordings of 20 renowned jazz pianists and discovered that harmonic progressions, rhythmic structures, and melodic motifs contribute to the distinctiveness of each performer's style.
The models, trained on MIDI-transcribed performances, achieved up to 94.4% accuracy in identifying the pianists. An interpretable model further refined the analysis, correctly attributing performers 91.3% of the time by separating the contributions of melody, harmony, rhythm, and dynamics. Harmonic features proved to be the most informative, followed by rhythm and melody, while dynamics showed the least accuracy.
While the approach demonstrates the potential for exploring musical patterns that make performers distinct, the authors acknowledge limitations in capturing timbral qualities such as vibrato and pitch bending. Future research could expand the methodology to other genres, instruments, and underrepresented musicians, providing valuable insights into artist attribution, cultural heritage, and music education.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
