Cleansed PHAT GCC based sound source localization

Abstract

Sound source localization (SSL) is one of the important techniques used in various engineering fields such as monitoring and surveillance system, and so on. When SSL is utilized in robot auditory system, it is essential for Human-Robot Interaction (HRI) in that robot can recognize its working environment and what direction a speaker is locating. Generalized cross correlation (GCC) function has been frequently used for SSL due to its fast computation and less requirement of high system resources. Artificial ear which is the robot auditory hardware we are dealing with consists of 2 microphones and a single pinna structure. This is locating on the both sides of robot head where free field condition can't be applicable due to reflected wave caused by robot platform such as robot shoulder, etc. This paper has focused on modification of spatially mapped GCC function, which was proposed by B. Kwon, for application into artificial ear structure. We proposed cleansed PHAT GCC function for use in spatially mapped GCC method. Cleansed PHAT GCC function combines cleansing method and conventional PHAT GCC function for eliminating reflected wave which causes free field condition break down. Simulation results using proposed cleansed PHAT GCC function showed that 93% SSL performance is obtainable when a single sound source exists.