Perceptual evaluation of a multiband acoustic crosstalk canceler using a linear loudspeaker array

Abstract

We propose and evaluate an acoustic crosstalk canceler for binaural sound presentation based on an 8-channel linear loudspeaker array. The system uses traditional LSFI beamforming and path estimation in the mid and low frequency range, respectively. Simulations show a channel separation of around 40 dB over the vast part of the considered frequency range. The channel separation in a real-world implementation was measured to be 10 dB – 25 dB over a broadband spectrum of 250 Hz – 9000 Hz. When comparing localization precision of virtual binaural signals to ground-truth presentation with headphones, a 21 subject study revealed significantly higher localization precision (F(1) = 23.56; p < 0.001) when discounting front-back confusions, which in turn occurred significantly more often for most participants. Perceptually, externalization was rated significantly higher (t(20) = 3.983; p > 0.001); task difficulty, timbre and spaciousness where rated the same accross both presentation methods. Robustness of the method was high and no fixation of the subjects' heads was applied.