Research proposal: Problem Context, Limitations + Research Gap, Formulate Core Problem Statement

[SteveAndCow]

Overview:

Status Quo
Proposals from Papers
Tradeoffs
Our Proposal
Can use the notebook LM: https://notebooklm.google.com/notebook/0ca55de8-13bb-4bbe-b3f9-9c517053ec5a

[SteveAndCow]

Model Specialization: Lu et al. (2025) suggest designing different classification structures for different categories of music, indicating a gap in achieving a truly universal MSA model that performs optimally across diverse genres (e.g., pop vs. classical).

For example, since a few of us have interest in electronic music, and in initial tests of using SongFormer, the section labels aren't necessarily relevant enough to the genre -- i.e. labelling the buildup section of a DNB song as "prechorus, chorus".

[25ohms]

Issues:

• Genres tend to be mislabeled, heuristic method for determining Genres into mutually exclusive bucket.

Potential Solution:

• Focus on BPM as our method of categorization rather than Genre. We can create more diverse datasets.