Research Proposal: Discovering Motives in Multi-track Music Data for Music Representation Learning

Research | 2023 (Proposal submitted)

Advance music modeling by integrating and refining motif discovery techniques.
Propose annotations for the BPS-motif dataset, our lab's latest symbolic motif and occurrence dataset.

Abstract

Uncovering Motifs on multi-track symbolic music data is relatively under-explored and remains challenging. Our proposal addresses challenges in musical motif discovery by contributing to the BPS motif, the latest symbolic motif and occurrence dataset. We will consider a data-driven motif-discovery algorithm, derived from MIR tasks on melody extraction, boundary detection, clustering, and multitasking learning. Drawing on music theory and employing advanced deep-learning techniques, we aim to offer valuable insights for music modeling and generation.

Figure 1. Overview of our data-driven motif-discovery pipeline

>> Proposal