Uncovering audio patterns in music with Nonnegative Tucker Decomposition for structural segmentation

TL;DR

This paper demonstrates that Nonnegative Tucker Decomposition effectively uncovers musical patterns and aids in structural segmentation of pop songs, outperforming some state-of-the-art methods without extensive training.

Contribution

It introduces the application of Nonnegative Tucker Decomposition to identify musical motifs and segment structures in audio data, offering a new approach to music analysis.

Findings

NTD captures repeated motifs as linear combinations of patterns

Features derived from NTD improve structural segmentation accuracy

Results challenge existing example-based learning approaches

Abstract

Recent work has proposed the use of tensor decomposition to model repetitions and to separate tracks in loop-based electronic music. The present work investigates further on the ability of Nonnegative Tucker Decompositon (NTD) to uncover musical patterns and structure in pop songs in their audio form. Exploiting the fact that NTD tends to express the content of bars as linear combinations of a few patterns, we illustrate the ability of the decomposition to capture and single out repeated motifs in the corresponding compressed space, which can be interpreted from a musical viewpoint. The resulting features also turn out to be efficient for structural segmentation, leading to experimental results on the RWC Pop data set which are potentially challenging state-of-the-art approaches that rely on extensive example-based learning schemes.