Date of Award

Spring 5-3-2024

Document Type

Thesis (Master's)

Department or Program

Computer Science

First Advisor

Michael Casey

Second Advisor

Lorie Loeb

Third Advisor

William Cheng


Generative music, first introduced by composers like Brian Eno and David Cope in the mid-to-late 20th century, has evolved through many stages, with diverse applications across music and technology companies as well as the game industry. However, despite this widespread interest, there remains a notable lack of the foundational understanding of composition and individual expressiveness in current generative systems that was apparent in the work of early composers. This paper advocates for a shift towards prioritizing compositional thought in system design to foster greater diversity and innovation within generative music. To demonstrate this approach, a novel system synthesizing two distinct musical pieces into a single cohesive composition was developed and evaluated. The system’s efficacy was evaluated through two user tests involving 35 participants. In the first test, participants listened to excerpts from various video game composers as well as music generated by combining different pieces. They rated the resulting compositions based on valence and energy, and then discussed their emotional and thematic qualities. In the second test, participants compared music clips generated by different models, including the system developed in this study, rating them on compositional quality. Results exhibit the system’s effectiveness in blending thematic material from its two inputs and producing compositions of comparable quality to existing models. The success of this system underscores the importance of reintegrating musical expression alongside technological advancements in contemporary generative music research and industry practices.