Speaker tracking based on distributed particle filter and interacting multiple model in distributed microphone networks

Abstract

The reverberations and Gaussian noises from indoor environments pollute speech signals and bring the degradation of speaker tracking performance. In the paper, a speaker tracking method based on the distributed particle filter (DPF) and the interacting multiple model (IMM) is proposed in distributed microphone networks. The generalized cross-correlation (GCC) function is first employed to estimate the time difference of arrival (TDOA) of speech signals received by two microphones at each node. To overcome the adverse effects of reverberation and noise, multiple TDOAs are selected to calculate the multiple-hypothesis model as the local weight of the DPF. To simulate a real speaker motion, the IMM algorithm is applied and a calculation method of the local measurement is presented based on multiple TDOAs. Finally, the speaker tracking method based on the DPF and IMM is performed to track a moving speaker and obtain a global consistent position estimate. The proposed method can track the moving speaker in reverberant and noisy environments with high tracking accuracy in distributed networks, and it is robust against the fault nodes. Simulation results demonstrate the validity of the proposed speaker tracking method.