A COMBINED CASCADING SUBSPACE AND ADAPTIVE SIGNAL ENHANCEMENT METHOD FOR STEREOPHONIC NOISE.pdf

A COMBINED CASCADING SUBSPACE AND ADAPTIVE SIGNAL ENHANCEMENT METHOD FOR STEREOPHONIC NOISE REDUCTION T. Hoya, A. Cichocki, T. Tanaka, G. Hori, T. Murakami , and J. A. Chambers
Laboratory for Advanced Brain Signal Processing, BSI RIKEN, 2-1, Hirosawa, Wakoh-City, Saitama 351-0198, Japan Department of Electronics and Communication Engineering, Meiji University, 1-1-1, Higashi-mita, Tama-ku, Kawasaki, Japan
Centre for Digital Signal Processing, Division of Engineering, King’s College London, WC2R 2LS, U.K. ABSTRACT A novel stereophonic noise reduction method is pro- posed based upon a combination of cascaded subspace fil- ters, with delay and advancing elements alternatively in- serted between the adjacent cascading stages, and two-channel adaptive signal enhancers. Simulation results based upon real stereophonic speech contaminated by two correlated noise components show that the proposed method gives im- proved enhancement quality, as compared to conventional nonlinear spectral subtraction approaches, in terms of both segmental gain and cepstral distance performance indices. 1. INTRODUCTION In the last few decades, noise reduction has been a topic of great interest in speech enhancement. One of the classical and most commonly used methods is based upon nonlinear spectral subtraction (NSS) [1]. In NSS methods, however, due to the block processing based approach, it is well known that such methods introduce annoying artifacts, which are often referred to as undesirable “musical tone”, in the en- hanced speech. Moreover, in many cases, such methods also remove some speech components in the spectra which are fundamental to the intelligibility of the speech. This is a particular problem at lower SNRs. The performance is also quite dependent on the choice of many parameters, such as, spectral subtraction floor, over-subtraction factors, or over-subtraction corner frequency parameters. The opti- mal choice of these parameters in practice is therefore very difficult