Isotropic Noise Suppression in the Power Spectrum Domain by Symmetric Microphone Arrays

In this paper, we propose a new array processing framework for suppressing isotropic noise in the power spectrum domain. As a theoretical basis, we discuss the characteristics of the isotropic noise covariance matrix and show that it can be diagonalized by a definite unitary matrix when the microphone array has a certain symmetry. By the diagonalization, our method gathers noise components to diagonals of the basis-transformed covariance matrix and restores the power spectrum of the target source in a specified direction from non-diagonal components by Maximum Likelihood (ML) method. We performed simulations which show the efficiency of our method for both stationary noise field and non-stationary noise field situations.