Histogram based DOA estimation for speaker localisation in reverberant environments

This paper introduces a Direction of Arrival (DOA) estimation technique for microphone arrays that has improved robustness in reverberant environments. The technique applies a separate beamformer at each frequency to estimate the DOA of the speech signal at that frequency. A histogram of the DOA estimates from all frequencies is then formed, which represents the probability of the signal coming from any particular direction. The histograms can be averaged over consecutive data blocks to improve the reliability. The peak of this angular probability distribution gives an estimate of the DOA. This technique can be used to determine the probability of the person being at any particular location in the room, by combining the angular probability distributions from multiple microphone arrays at different locations in the room.