3-D Multiple Sound Source Localization Based on a Five-Element Microphone Array

Abstract

A 3-D multiple sound source localization (SSL) method is proposed in this work, which uses independent vector analysis (IVA) combining with an elaborately designed five-element microphone array. The proposed method separates the source signals by IVA. Then for each separated signal, four time-difference-of-arrival (TDOA) values are obtained. With the five-element microphone array, the four TODAs are used to realize the localization of each source separately by analytical solution. To meet the prerequisite of IVA, the properties of the mixing matrix with the five-element microphone array are analyzed and studied for 2 and 3 sound sources. And it is proved that, the configuration of the five-element microphone array can avoid the ill condition where the mixing matrix at each frequency bin has linearly dependent columns. To reduce the computation cost of IVA, the microphones with the same number of the sound sources are selected from the array, and their signals are used for audio source separation. Meanwhile, considering the calculation stability, the selected microphones are required to minimize the condition number of the microphone signal covariance matrix. To investigate the effectiveness and the localization performance of the proposed method, a practical five-element microphone array is used and 3-D multiple SSL experiments are carried out. The TDOA values are obtained by the generalized cross correlation based on the phase transform (GCC-PHAT). The experimental results show that the proposed method is effective and the maximum root mean square error of localization is less than 3 cm . Compared with the conventional methods, the proposed method has the advantages of lower computation cost and fewer microphones, and can locate sources close to each other.