Acoustic Source Tracking in Reverberant Environment Using A Novel PSO Particle Filter Framework

In this paper, several methods are combined together and applied to a Particle Filter based on Particle Swarm Optimization acoustic source tracking framework in order to solve some problems of sound source tracking. The Time Difference of Arrival (TDOA), which is extracted from the audio signal received by the microphone pair, makes observations. The Langevin Model is used to describe the speaker's motion characteristics. The pseudo-likelihood function is constructed by using the Global Coherence Field (GCF) function to update the particle weight, in order to solve problem of the phenomenon of spurious source due to noise and reverberation. According to the iterative process and particle flight conditions, the inertial weights of the particles are adjusted nonlinearly and dynamically. The concepts of superior flight speed and inferior flight speed are introduced to Particle Swarm Optimization, which helps jump out of local optimum and solves the problem of lack of particle diversity. Thus, the faster tracking of sound source is realized. In the framework, with fewer particles, the robustness of tracking is enhanced and accuracy of tracking is improved. Finally, the effectiveness of this framework for acoustic source tracking is verified through several Monte Carlo simulation experiments.