Institute of Information Science, Academia Sinica

Recent advances in music informatics have opened up various possibilities in how we create, play, and enjoy music. However, most of this research has focused on Western pop and classical music. In this talk, I introduce several works on applying music informatics to Korean traditional music.

These works cover computational musicology, such as finding tori (the modal categories of Korean folk songs) in about 700 hours of field recordings and classifying modes in pansori (a genre of sung storytelling) using recordings labeled by a pansori expert. They also cover generative tasks: reviving 15th-century court music through symbolic music generation, and our ongoing work on pansori audio generation. The court music survives only as a simple melody, and our transformer models turned it into a six-part ensemble arrangement, which was performed by the Court Music Orchestra of the National Gugak Center. Lastly, I present a new approach to a challenging task: analyzing pitch contours as discrete symbols. Korean traditional music uses many continuous pitch movements, such as vibrato and sliding, that are hard to write as notes. We used VQ-VAE tokenization to learn a vocabulary of short pitch-contour patterns without labels, and some of the learned patterns matched known ornament types. I also discuss how we, as music informatics researchers, can approach Korean traditional music differently in the future and what we can contribute.

Dasaem Jeong is an associate professor in the Department of Art & Technology at Sogang University and leads the Music & Arts Learning (MALer) Lab. He received his Ph.D. from KAIST and has been at Sogang since 2021. He has long researched Korean traditional music and Western classical music, and is actively working on music information retrieval and computational musicology. He also served as a general chair of ISMIR 2025.