Sound source localization by microphone array on a mobile robot using eigen-structure based generalized cross correlation

Sound source localization is an important feature in robot audition. This work proposes an eigenstructure-based generalized cross correlation method for estimating time delay between microphones. From experiments, it shows that this method is more robust as compared with conventional GCC using PHAT weighting. By knowing the delays among microphones, both the sound source location and sound speed can be estimated using an array. Further, the correctness of sound speed estimation can be utilized to determine locations of multiple sources. This work derives in detail the method to estimate the directions of dual sources. A digital microphone array and associated acquisition system is developed for conducting the experiment. Experimental results prove the effectiveness of the proposed algorithms.