Untangling Highly Correlated Vocals in Multiple Singing Voices Separation

Anonymous Authors
Affiliation withheld for review

Abstract

Multiple singing voice separation (MSVS) is closely related to speech separation, yet presents greater challenges due to the highly correlated nature of singing voices and the scarcity of multiple singing datasets. Existing studies on MSVS can be broadly classified into two categories: choral music separation and popular music separation. The latter remains underexplored and continues to exhibit limited performance. In this work, we address these limitations by introducing: (1) a data mining strategy for constructing highly correlated training mixtures, (2) a reverse attention mechanism to suppress highly correlated regions between outputs, and (3) a magnitude penalty loss that penalizes spectrogram regions containing energy that should exclusively belong to the other output. Experimental results demonstrate that our approach achieves substantial performance gains over prior methods.

Audio Samples

Medleyvox Duet Samples

🎵 MedleyVox Duet Sample1

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

🎵 MedleyVox Duet Sample2

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

🎵 MedleyVox Duet Sample3

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

🎵 MedleyVox Duet Sample4

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

Medleyvox Unison Samples

🎵 MedleyVox Unison Sample1

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

🎵 MedleyVox Unison Sample2

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

🎵 MedleyVox Unison Sample3

Mixture

Ground Truth

MedleyVox

TIGER

Proposed

Pop Samples

🎵 Pop Sample1: Jason Mraz and Colbie Caillat - Lucky

Mixture

MedleyVox

TIGER

Proposed

🎵 Pop Sample2: Shawn Mendes and Camila Cabello - Señorita

Mixture

MedleyVox

TIGER

Proposed

🎵 Pop Sample3 : Lady Gaga and Tony Bennett - I’ve Got You Under My Skin

Mixture

MedleyVox

TIGER

Proposed

🎵 Pop Sample4 : Verandah Project (Dong Ryul Kim and Sang-soon Lee) - Bike Riding

Mixture

MedleyVox

TIGER

Proposed