Artificial Intelligence (AI) can do many things, but using AI to generate music has been an ongoing challenge. An even more difficult task is the creation of musical pieces that match human-specific preferences.
New research from Technion M.Sc. students Nadav Bhonker (already graduated) and Shunit Haviv Hakimi, and their advisor Professor Ran El-Yaniv of the Henry and Marilyn Taub Faculty of Computer Science at the Technion-Israel Institute of Technology, indicates that it is possible to model and optimize personalized jazz preferences.
To develop personalized preferences, Bhonker and Haviv Hakimi, both amateur jazz musicians, first introduced the AI to hundreds of original jazz solos performed by saxophone giants including Charlie Parker, Stan Getz, Sonny Stitt, and Dexter Gordon, as well as data on user preferences. Bhonker and Haviv Hakimi were then able to extract individual preference models developed by the listening habits of several human listeners.
The team also conducted a plagiarism analysis to ensure that the generated solos are unique, rather than a concatenation of phrases previously seen in other jazz solos.
The team found that while the AI-generated solos are often interesting or pleasing to listen to, they lack the qualities of professional jazz solos, such as motif development or variations. Prof. El-Yaniv hopes to overcome this challenge in future research, possibly by introducing a larger dataset for the AI to learn from.
Bhonker, Haviv Hakimi, and Prof. El-Yaniv presented their findings in a paper titled “BebopNet: Deep Neural Models for Personalized Jazz Improvisation.” It was presented during the 21st International Society for Music Information Retrieval Conference.