APSIPA Transactions on Signal and Information Processing > Vol 14 > Issue 1

Music Bleeding-sound Reduction Based on Time-channel Nonnegative Matrix Factorization

Yusaku Mizobuchi, National Institute of Technology, Kagawa College, Japan, Daichi Kitamura, National Institute of Technology, Kagawa College, Japan, kitamura-d@t.kagawa-nct.ac.jp , Tomohiko Nakamura, The University of Tokyo, Japan, Norihiro Takamune, The University of Tokyo, Japan, Hiroshi Saruwatari, The University of Tokyo, Japan, Yu Takahashi, Yamaha Corporation, Japan, Kazunobu Kondo, Yamaha Corporation, Japan
 
Suggested Citation
Yusaku Mizobuchi, Daichi Kitamura, Tomohiko Nakamura, Norihiro Takamune, Hiroshi Saruwatari, Yu Takahashi and Kazunobu Kondo (2025), "Music Bleeding-sound Reduction Based on Time-channel Nonnegative Matrix Factorization", APSIPA Transactions on Signal and Information Processing: Vol. 14: No. 1, e18. http://dx.doi.org/10.1561/116.20250020

Publication Date: 21 Jul 2025
© 2025 Y. Mizobuchi, D. Kitamura, T. Nakamura, N. Takamune, H. Saruwatari, Y. Takahashi and K. Kondo
 
Subjects
Independent component analysis,  Audio signal processing,  Enhancement,  Multidimensional signal processing,  Sensor and multiple source signal processing,  Signal decompositions,  Source separation,  Statistical signal processing,  Statistical/Machine learning
 
Keywords
Blind source separationbleeding soundnonnegative matrix factorizationmaximum a posteriori estimation
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 46 times

In this article:
Introduction 
Conventional Methods 
Proposed Method 
Experimental Results and Discussion 
Initialization Robustness Analysis 
Conclusion 
References 

Abstract

When we place microphones close to a sound source near other sources in audio recording, obtained audio signals include undesired sound from the other sources, which is often called bleeding sound. For many audio applications including onstage sound reinforcement and sound editing after a live performance, it is important to reduce the bleeding sound in each recorded signal. However, since microphones are spatially apart from each other, typical phase-aware blind source separation (BSS) methods cannot be used. We propose a phase-insensitive method for blind bleedingsound reduction. This method is based on time-channel nonnegative matrix factorization, which is a BSS method using only amplitude spectrograms. In the proposed method, a gamma prior distribution is introduced for the frequency-wise leakage gains of the bleeding sound component to estimate the mixing matrix. This estimation can be interpreted as maximum a posteriori probability estimation. From the experimental results, it is confirmed that the proposed method can reduce the music bleeding sound with higher accuracy than the other methods in both simulated and real situations. It is also confirmed that the proposed method achieves robust performance against parameter initializations, which is an important advantage in practical applications. The reason of this robustness is experimentally revealed.

DOI:10.1561/116.20250020