Machine learning of artistic fingerprints in jazz

Problem — This work addresses the gap in automated identification of jazz pianists from audio recordings, a task that has not been extensively explored in the literature. The study is a preprint and has not undergone peer review.

Method — The authors develop a machine learning pipeline that analyzes audio features such as melody, harmony, rhythm, and dynamics to extract individual musical fingerprints of 20 iconic jazz pianists. The specific architecture and training compute details are not disclosed in the available text.

Results — The model achieves up to 94% accuracy in identifying the pianists from audio recordings, although the available text does not report quantitative results against specific baselines or benchmarks.

Limitations — The authors do not specify limitations in their methodology, but potential issues could include the generalizability of the model to other genres or the need for a larger dataset to improve robustness.

Why it matters — This research has implications for music information retrieval and the study of musical styles, potentially influencing future work in automated music analysis and classification, as published in Nature Machine Intelligence.