Improving the Classification of Rare Chords with Unlabeled Data

TL;DR

This paper investigates methods to enhance the recognition of rare chords in automatic chord recognition by applying focal loss and self-learning techniques, with the latter showing the most significant improvement.

Contribution

It introduces the adaptation of self-learning with noise addition for rare chord classification, outperforming other methods in the domain.

Findings

Self-learning with noise addition improves rare chord recognition.

Focal loss alone provides some improvement but less than self-learning.

Combining both methods yields further gains.

Abstract

In this work, we explore techniques to improve performance for rare classes in the task of Automatic Chord Recognition (ACR). We first explored the use of the focal loss in the context of ACR, which was originally proposed to improve the classification of hard samples. In parallel, we adapted a self-learning technique originally designed for image recognition to the musical domain. Our experiments show that both approaches individually (and their combination) improve the recognition of rare chords, but using only self-learning with noise addition yields the best results.