Sound localization and tracking using distributed microphones fusion: Maximum Likelihood or Maximum A-Posteriori approach?

Sound source localization and tracking of moving sound source (displacement tracking) has received a lot of attention lately because of its importance in a number of next generation applications. In this paper we perform these operations using distributed microphone arrays. We generate spatial likelihood function using steered response power with phase transform filter (SRP-PHAT) for all microphone arrays and produce a unified fused spatial likelihood function (SLF). We show that the localization estimates from fused SLF is based on the maximum likelihood criteria. We find that using the maximum likelihood criteria requires exhaustive computations and is susceptible to noise and reverberation. We show that by modeling the a-priori information and using it for localization estimates under maximum a-posteriori criteria, results are improved in terms of computational load even under reverberant conditions. We model the a-priori information using an adaptive trajectory estimation procedure while taking into account the geometrical constraints of the room. Last section describes the simulation results for the procedure described in this paper.