Listener-Position Adaptive Crosstalk Cancelation Using a Parameterized Superdirective Beamformer

摘要

We present a method for cancelation of the acoustic crosstalk between the ears of a listener using a linear loudspeaker array. This allows for controlling the signals that arise at listener's ear independently of each other so that binaural cues can be imposed. We employ superdirective near-field beamforming (SDB) over the vast part of the frequency range. Allowing the listener to move in front of the array requires the computation of a new SDB solution for each possible position of the listener, which is a computationally expensive procedure. We show that the beamformer weights exhibit a very smooth evolution for listening positions along a line parallel to the array. This makes it straightforward to parameterize pre-computed beamformer weights with a few parameters for each frequency. Upon realtime execution, the beamformer weights can be determined efficiently for any arbitrary position along the listener contour from the parameters with negligible error. Simulation results show that the proposed beamformer provides a higher channel separation than previously published solutions while maintaining robustness.