Complex ratio masking for singing voice separation

TL;DR

This paper introduces a complex ratio masking approach using DenseUNet with self-attention for singing voice separation, leveraging phase information to significantly improve separation quality over existing methods.

Contribution

It presents a novel complex ratio masking method that estimates both real and imaginary parts of STFT, incorporating self-attention and ensemble techniques for enhanced separation performance.

Findings

Outperforms recent state-of-the-art models in voice separation

Utilizes phase information for better separation quality

Employs DenseUNet with self-attention for accurate complex STFT estimation

Abstract

Music source separation is important for applications such as karaoke and remixing. Much of previous research focuses on estimating short-time Fourier transform (STFT) magnitude and discarding phase information. We observe that, for singing voice separation, phase can make considerable improvement in separation quality. This paper proposes a complex ratio masking method for voice and accompaniment separation. The proposed method employs DenseUNet with self attention to estimate the real and imaginary components of STFT for each sound source. A simple ensemble technique is introduced to further improve separation performance. Evaluation results demonstrate that the proposed method outperforms recent state-of-the-art models for both separated voice and accompaniment.