University of Twente Student Theses


AI-Composed Music for User Preference Using Reinforcement Learning

Mysliwiec, D. (2023) AI-Composed Music for User Preference Using Reinforcement Learning.

[img] PDF
Abstract:Artificial intelligence has become a powerful tool in automatic music generation and music recommendation systems thanks to the field's rapid development. Nonetheless, few solutions have been proposed for easily generating music based on a non-expert user's preference. The results of an extensive literature review show that the main issue to combat was that of subjectivity in music, and gave an indication that a reinforcement learning algorithm combined with a deep learning model for user preference could be an appropriate solution. Based on these findings a new automatic music composition system that relies on a reinforcement learning algorithm and models preferences based on the user's ratings of transformer-generated pieces, improving tailoring to the user over iterations of the algorithm. The system was evaluated through human interaction and shows promising results with a 44.7\% improvement to the mean user rating of generated pieces in 6 iterations of the algorithm. The rating prediction model also achieves high accuracy with a difference between the predicted and received ratings of just 0.81 on a 10-point scale. Because the system relies on no prior knowledge, with such effectiveness it could give access to music compositions for user preference to a wider non-expert audience.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page