Tracking and Voice Separation of Moving Speakers based on IMM-PDA filters

The problem addressed in this work is that of separating the voices of simultaneously active moving speakers using a single microphone array. We adapt the multi-sensor multi-target tracking theory to the context of microphone arrays, in order to form a receptive beam that locks on each moving speaker on an extended time basis and, therefore, achieves voice separation. Our approach (a) incorporates kinematical information of speakers’ movement by using an interacting multiple model (IMM) estimator per speaker in order to constrain the evolution of Direction of Arrival (DOA) measurements, and (b) can directly account for measurement origin uncertainty by using the probabilistic data association (PDA) technique in conjunction with the IMM estimator. Finally, we demonstrate the function of the gate as a means to initiate and terminate track segments corresponding to phrases.