A Multi-speaker Tracking Approach under Reverberation Environment based on Finite Set Theory

This work studies the multi-speaker tracking in the case of high reverberation. To remove the spurious peaks caused by the reverberations, a density clustering concept is utilized to create peak clusters based on steered-response power (SRP) values. After that, the location information is fed into a tracking algorithm, where probability hypothesis density (PHD) filter is developed. In the process of PHD, the SRP values are also exploited to better decide the update of the states. From numerical studies, together with clustering, the proposed tracking algorithm can effectively remove the influence of reverberations and produce a superior tracking performance.