Source localization algorithms in binaural processing models

Abstract

This thesis presents sound source localization algorithms in binaural processing models. We propose two new binaural source localization algorithms. One is a time-domain processing algorithm for finding sound source direction using principal component analysis (PCA) of input correlation matrix. And another is a frequency domain method to estimate useful binaural cues of interaural time difference (ITD) and interaural intensity difference (IID) using zero-crossings and peak amplitudes (ZCPAs) of cochlear filter output signals. Both of our methods are practical to the source localization problem. However the PCA method has relatively large variance for an accurate estimation of source direction and cannot be applied to multiple source conditions. The ZCPA method can capture accurate source directions even in multiple sound sources. With simulations in diffusive and interfering noisy conditions, we verified that the suggested algorithms are effective to the source localization problem.