馃幎How can we learn to follow along to a music performance in a score without direct supervision?馃幎
Introducing FuSiLi (Fused Sinkhorn-Localized Similarity)!
A multimodal contrastive learning approach that learns precise local alignments using only global supervision.
馃У
We train with zero local alignment labels and evaluate across PDMX, MSMD, and real-world out-of-domain string quartets (YTSV).
Dive into the full paper, code, and demo here:
馃搫 Paper: arxiv.org/abs/2607.10023
馃捇 Code: github.com/irmakbky/fusili
馃幑 Demo: irmakbky.github.io/fusili/
6/
FuSiLi will appear at ISMIR 2026! Huge thanks to all my collaborators @zacknovack @sake_min @DasaemJ @chrisdonahuey
7/7
4
268
Irmak Bukey retweeted
Introducing 馃幖馃MulTTiPop, an evaluation dataset of multitrack pop music for automatic music transcription!
MulTTiPop contains 3.5 hours (572 segments) of aligned multitrack pop audio and MIDI, and can be previewed at gclef-cmu.org/multtipop
馃У1/7
馃幖 Can we turn music into readable code?
Excited to share Decomposer, a post-training framework for symbolic music decompilation!
Given MIDI, Decomposer recovers an executable program in Strudel that reconstructs the music while exposing patterns, harmony, rhythm, and voices as editable code.
馃У [1/5]


