Single sound source localization using multi-layer perceptron

Abstract

A localization of single sound source is investigated in this paper. The aim of investigation was to propose a technique for single sound source localization based on sound level differences in microphone array. A rectangular array of four microphones was used for sound capturing from single sound source. Location of the sound source is estimated by sending average amplitude of the signal, received from four microphones to a Multi-Layer Perceptron. Different structures of the MultiLayer Perceptron were tested on simulated sound source locations and on the real measured signals. Estimated relative sound source position prediction error was as low as 4 % in computer simulation and 1.9 % in practical experimentation. Investigation results presented in this paper shows, that moderate precision of single sound source localization may be received using a MultiLayer Perceptron with only signal amplitude based features sent to the input of the network.