Particle methods for real-time sound source localization based on the Multiple Signal Classification algorithm

Abstract

Multiple Signal Classification (MUSIC) is a microphone-array signal processing method that achieves high resolution for the estimation of acoustic directions of arrival (DOA)[1]. However, MUSIC in its original form does not estimate the distance from the source to the microphone array. In this work, we propose a method to conduct two-dimensional sound source localization using multiple pairs of microphones. The idea of sampling by particles is combined with MUSIC, and the method has been implemented to run in real-time. To improve the performance against reverberation, the result of localization can be post-processed by Kalman filters or particle filters so the location of the source is continuously tracked. Because the new method allows microphone arrays to process their input in parallel, it is potentially suitable to be deployed to a sensor-network platform.