Binaural bearing only tracking of stationary sound sources in reverberant environment

Abstract

In this work we present a framework for the estimation of the Cartesian position of stationary sound sources in reverberant environments and under the influence of heavy clutter based on binaural bearing measurements. We employ a particle filter (PF) on binaural measurements to estimate the position of the sound sources in a bearing only tracking (BOT) formulation and investigate how the estimation accuracy can be improved in reverberant environments by applying a gating method that is inspired by the precedence effect. We evaluate the interaural coherency in order to identify time frequency units of the received signals that show a high linear dependency and therefore are potentially dominated by the direct sound emitted by sound sources. We use a particle filter for state estimation and lay out the theoretical model for state representation, propagation and estimation. The feasibility of the presented methods is evaluated in simulations and we give first results of tracking performance when applied to real world binaural localization measurements of a sound source in a typical reverberant scenario. Our results show that gating the binaural bearing measurements with the interaural coherency can improve localization accuracy to a large degree.