Co-Creative Music Generation through Interactive Preference Estimation

We study co-creative music generation that estimates users’ affective preferences on the fly through dialogue and reflects them in diverse creative suggestions. Using In-Context Learning and In-Context Preference Learning, we take in value criteria without additional training and aim to support humans and AI in collaboratively creating new musical expressions.

Related papers:

• Futa Hidaka, Naomi Imasato, Kazuki Miyazawa, Takato Horii. “Creative Music Generation via Value-Criterion Learning Using In-Context Learning” (in Japanese). Annual Conference of the Japanese Society for Artificial Intelligence (JSAI), May 2025.
• Makoto Teshirogi, Futa Hidaka, Yuichiro Yoshikawa, Takato Horii. “Co-Creative Music Generation Model with In-Context Preference Learning for Acquiring Affective Expressions and Diverse Suggestions” (in Japanese). Annual Conference of the Japanese Society for Artificial Intelligence (JSAI), June 2026.