Estimating Power Spectral Density for Acoustic Signal Enhancement - an Effective Approach for Practical Applications

Abstract

The acoustic signal enhancement has been still a challenging problem especially when a method has to be effective and feasible in practical applications although various approaches have been studied for many decades. Applying a postfilter such as the Wiener filter, which manipulates the spectral amplitude of a signal, is known to be an effective approach for the practical applications. However the power spectral densities (PSD) of sound sources need to be estimated beforehand for deriving the postfilter. In this talk a method for estimating the PSD of sound sources located in spatially different positions is introduced. The method estimates the PSD of each sound source from the observation of multiple beamformers (or more simply multiple directional microphones). After introducing the principle of the method several practical applications that utilise the proposed PSD estimation method will be presented with some experimental results demonstrating the potential of the proposed method for solving challenging practical problems. The talk will also refer to a few cases where the proposed method was applied to other acoustic applications such as blind estimation of the direct-to-reverberation ratio.