Multi-array multi-speaker tracking

Abstract

In this work we deal with the fusion of the estimates of independentmicrophone arrays to produce an improved estimate of the Direction of Arrival(DOA) of one moving speaker, as well as localization coordinates of multiple moving speakers based on Time Delay Of Arrivals (TDOA). Our approach (a) fuses measurements from independent arrays, (b) incorporates kinematic information of speakers' movement by using parallel Kalman filters, and (c) associates observations to specific speakers by using a Probabilistic Data Association (PDA) technique. We demonstrate that a network of arrays combined with statistical fusion techniques provides a consistent and coherent way to reduce uncertainty and ambiguity of measurements. The efficiency of the approach is illustrated on a simulation dealing with beamforming onemoving speaker on an extended basis and localization of two closely spaced moving speakers with crossing trajectories.