Multiple speaker tracking with the GLMB filter

Abstract

© 2017 IEEE. In this paper we propose a new solution to the problem of tracking multiple speakers from multiple microphone arrays in a reverberant acoustic environment. The acoustic environment with its complex reflection patterns with its underlying data association uncertainty pose the two most significant challenges in the multi-speaker tracking problem. We provide an approach that employs individual Time Difference of Arrival measurements collected by pairs of microphones in using multiple distributed pairs in conjunction with the Generalized Labeled Multi-Bernoulli (GLMB) tracker. The distributed measurements together with the GLMB tracking filter exploits the spatiotemporal correlation of the true sources from data frame to data frame, whereas the spurious measurements arising from reverberations exhibit no temporal consistency as the speakers move in the room.