EUSIPCO 2025: A new research presentation was given

In September 2025, Shun Sawada presented the following research at the European Signal Processing Conference (EUSIPCO 2025), held in Sicily, Italy.

Shun Sawada, “Contrastive Audio-MIDI Learning for Symbolic-domain Musical Instrument Classification,” In Proceedings of EUSIPCO 2025, pp.426–430, September 2025.

This study addresses the task of estimating musical instruments from symbolic music data such as MIDI. MIDI data contains performance information such as pitch and rhythm, but it does not directly include timbral information in the way audio signals do. To address this limitation, we aimed to incorporate instrument-related cues derived from audio information into symbolic music models by learning correspondences between audio signals and MIDI data.

Investigating how “instrumentality” emerges from structural characteristics in symbolic music data is expected to contribute not only to instrument classification, but also to music generation, arrangement support, and the interpretation of music understanding models. We will continue to develop this line of research as a bridge between music information retrieval and creative media support.