EUSIPCO 2024: A new research presentation was given

In August 2024, Shun Sawada presented the following research at the European Signal Processing Conference (EUSIPCO 2024), held in Lyon, France.

Shun Sawada, “Symbolic-domain Musical Instrument Classification using Knowledge Distillation from Audio-teacher to Symbolic-student,” In Proceedings of EUSIPCO 2024, pp.191–195, August 2024.

This study addressed musical instrument classification using symbolic music data such as MIDI. Symbolic music data contains information such as pitch, rhythm, and harmonic structure, but it does not directly include the timbral information found in audio signals. To address this limitation, we investigated a method for incorporating instrument-related cues derived from audio information into symbolic music models using a Knowledge Distillation framework, in which knowledge is transferred from a teacher model that takes audio signals as input to a student model that takes symbolic music data as input.

This study is part of our broader research toward modeling “instrumentality” in symbolic music data. In future work, we will analyze which elements—such as pitch range, rhythm, harmony, and voice structure—contribute to instrument estimation, and further develop this research toward performance understanding and creative support using music AI.