Multiple Sound Source Tracking Based on Generative Modeling and Recursive Bayesian Filtering of Spatial Gradient Spectra

This paper addresses the problem of tracking the positions of multiple sound sources using a compact-sized microphone array. Based on a generative model of the spatial gradient spectra of acoustic point sources, we propose incorporating the temporal continuity of source positions into the sequential posterior inference under a recursive Bayesian filter formulation. Simulation experiments with data generated by the image source model confirmed that the proposed idea had a certain effect on improving the accuracy of tracking moving sources.