Robust Minimum Variance Distortionless Response Beamformer based on Target Activity Detection in Binaural Hearing Aid Applications

In this work, a new beamforming algorithm for binaural hearing aids is introduced, with a good robustness to errors in the estimated target direction of arrival. The proposed design is based on a selection between a monaural MVDR beamformer and a binaural MVDR beamformer at each time-frequency bin. The decision of using either the monaural MVDR or the binaural MVDR at each time-frequency bin depends on a proposed target activity detection system, over an a priori determined target zone. The proposed target activity detection system uses a binaural beamforming method in target-cancelling mode, in order to get an estimate of the dominant speaker direction at each time-frequency bin. Simulations are performed using signals and propagation models obtained from real-life multichannel binaural hearing aids, to evaluate the resulting noise reduction and target distortion. Simulation results show the robustness of the proposed design to errors in the estimated target direction of arrival, in comparison with both the monaural MVDR and binaural MVDR algorithms.