[3O5-OS-22a] OS-22

[3O5-OS-22a-04] (OS invited talk) Harmonizing Music AI with Humans: Compositional Modeling and Controllability Enabled by Deep Learning

〇Taketo Akama¹ (1. Sony Computer Science Laboratories)

Keywords:Music Compositionality, Controllability, Low-resource Learning, Music and Brainwave, Music and Dance

In recent years, as the evolution of generative AI has garnered significant attention, a new era is emerging in the field of music where AI and humans collaborate and co-create. By leveraging music AI, individuals can focus on more value-added activities, while AI assists with complex processes and extensive tasks that have traditionally been challenging for humans. This collaboration is expected to enhance human capabilities. Furthermore, AI holds the potential to deepen our understanding of music, enhance our musical expressive abilities, and expand opportunities for the application of music, ultimately elevating the value of music.

What challenges must AI overcome to realize such a future? I believe that developing AI capable of understanding music compositionally, much like humans do, and interpreting human intent is crucial. In this pursuit, we model music as a hierarchical structure and construct systems that can be controlled not only with the compositional elements of music but also with brainwaves and dances. This approach aims to create AI that can grasp human intent in a multifaceted manner.

In particular, I will introduce a "generalizable approach that reduces dependence on human supervision" and invite you to explore the continuous development of music AI.