Nonnegative Tucker Decomposition with Beta-divergence for Music Structure Analysis of Audio Signals

TL;DR

This paper introduces a novel multiplicative updates algorithm for Nonnegative Tucker Decomposition using beta-divergence loss, specifically tailored for music structure analysis, demonstrating improved performance over Euclidean-based methods.

Contribution

It presents the first efficient implementation of NTD with beta-divergence loss for audio signals, enhancing pattern extraction in music information retrieval.

Findings

Beta-divergence based NTD outperforms Euclidean loss in music analysis.

Efficient tensor algebra implementation improves computational performance.

Unsupervised NTD with beta-divergence yields better music structure insights.

Abstract

Nonnegative Tucker decomposition (NTD), a tensor decomposition model, has received increased interest in the recent years because of its ability to blindly extract meaningful patterns, in particular in Music Information Retrieval. Nevertheless, existing algorithms to compute NTD are mostly designed for the Euclidean loss. This work proposes a multiplicative updates algorithm to compute NTD with the beta-divergence loss, often considered a better loss for audio processing. We notably show how to implement efficiently the multiplicative rules using tensor algebra. Finally, we show on a music structure analysis task that unsupervised NTD fitted with beta-divergence loss outperforms earlier results obtained with the Euclidean loss.