Nonnegative Tucker Decomposition with Beta-divergence for Music Structure Analysis of Audio Signals
This is an incremental improvement for music information retrieval researchers, offering a more efficient method for audio pattern extraction.
The authors tackled the problem of computing Nonnegative Tucker Decomposition (NTD) with beta-divergence loss for audio processing, proposing a multiplicative updates algorithm and showing it outperforms Euclidean loss in music structure analysis.
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.