Speech source localization based on copula rank correlation matrix

摘要

Direction-of-arrival (DOA) estimation of a speech source has generated great interest in recent years. Traditional methods, such as the steered response power (SRP) estimator, make use of a multi-element microphone array and can perform well in the presence of moderate noises and reverberations. Those algorithms are particularly good for Gaussian source signals, while for non-Gaussian speech signals, they may not be sufficient. For such cases, copulas can be useful to measure the dependence between random variables. In this paper, we present a copula-based source DOA estimator, based on a copula rank correlation matrix constructed from the rank correlation coefficient Kendall's τ, which we call CRCM. A set of simulations and real-data experiments demonstrates the advantages of the proposed estimator over the classical SRP method.