Spatial localization of sound sources: azimuth and elevation estimation

A model-based bio-mimetic sound localization method has been proposed for concurrent and continuous speech sources located in reverberant environment. However, the method only provides the estimation of azimuths of sound sources. The current paper is to extend the previous method to be available for sound sources existing in the 3D space by using a four microphone set. As the procedure of signal processing, arrival time differences of different microphone pairs are calculated at each available onset of sound components. The locus circles of different microphone pairs in the surface of Gaussian Sphere are then calculated and projected topographically to a horizontal plane, which consists the north pole, by the south pole projection. A two dimensional histogram in the projected plane corresponding to the source direction is created by summing all the projected maps of every onsets. Finally, the directions of sound sources are determined by the position of the major peaks in the two-dimensional histogram.