British scientists managed to train machine learning algorithms to identify jazz pianists based on their unique performance style. A convolutional neural network proved to be the most effective tool in this study.
This neural network demonstrated the ability to identify the performer in previously unseen recordings with an accuracy exceeding 94 percent. Furthermore, the development team created a model capable of analyzing compositional elements such as melody, harmony, rhythm, and dynamics in detail, which allows for an assessment of the influence of each of these components on the formation of the musician's individual musical signature.
The results of the work were presented in the scientific journal Nature Machine Intelligence. In addition to the research, the scientists developed a web application that allows users to listen to the distinctive features of the playing of each of the twenty studied pianists.