Speech Source Tracking based on Particle Filter under non-Gaussian Noise and Reverberant Environments

Abstract

Tracking a moving speech source in non-Gaussian noise environments is a challenging problem. A speech source tracking method based on the particle filter (PF) and the generalized correntropy function (GCTF) in non-Gaussian noise and reverberant environments is proposed in the paper. Multiple TDOAs are estimated by the GCTF and the multiple-hypothesis likelihood is calculated as weights for the PF. Next, predict the particles from the Langevin model for the PF. Finally, the global position of moving speech source is estimated in term of representation of weighted particles. Simulation results demonstrate the vadility of the proposed method.