Improving singing voice separation with the Wave-U-Net using Minimum Hyperspherical Energy

TL;DR

This paper enhances singing voice separation by integrating Minimum Hyperspherical Energy regularization into Wave-U-Net, leading to improved generalization and state-of-the-art results on the MUSDB18 dataset.

Contribution

It introduces MHE regularization to the Wave-U-Net architecture for singing voice separation, demonstrating improved performance over existing methods.

Findings

MHE regularization improves separation quality.

Achieves current best results on MUSDB18 dataset.

Enhances generalization of the Wave-U-Net model.

Abstract

In recent years, deep learning has surpassed traditional approaches to the problem of singing voice separation. The Wave-U-Net is a recent deep network architecture that operates directly on the time domain. The standard Wave-U-Net is trained with data augmentation and early stopping to prevent overfitting. Minimum hyperspherical energy (MHE) regularization has recently proven to increase generalization in image classification problems by encouraging a diversified filter configuration. In this work, we apply MHE regularization to the 1D filters of the Wave-U-Net. We evaluated this approach for separating the vocal part from mixed music audio recordings on the MUSDB18 dataset. We found that adding MHE regularization to the loss function consistently improves singing voice separation, as measured in the Signal to Distortion Ratio on test recordings, leading to the current best time-domain system for singing voice extraction.