Acoustic source tracking based on adaptive distributed particle filter in distributed microphone networks

In this paper, an adaptive distributed particle filter (ADPF) is proposed for single acoustic source tracking in distributed microphone networks (DMNs). To deal with spurious effects due to the reverberation and noise, a modified multiple-hypothesis model is first investigated by exploiting the generalized cross-correlation (GCC) function. Based on this model, the time-delay of arrival (TDOA) selection is performed for constituting the local observation. Then the acoustic source tracking is formulated as a Bayesian filtering problem under the assumption on the Langevin dynamic model of the source motion. Next, an adaptive distributed particle filter (ADPF) is presented to solve the Bayesian filtering problem for distributed acoustic source tracking. To improve the tracking performance, in the proposed ADPF, an adaptive and distributed computation method of the optimal proposal function is designed based on the Gaussian approximation, implemented by utilizing a Markov Chain Monte Carlo (MCMC) sampler and a consensus filter. The main advantage of the proposed acoustic source tracking method is the combination of the strength of the modified TDOA multiple-hypothesis model and the ADPF. Both simulation and real-world recording experiment results show that, the proposed ADPF has a relatively good tracking performance under different SNR conditions and reverberation environments.