Enhancing Spatial Post-Filters through Non-Linear Combinations

This paper introduces a method to enhance the spatial selectivity of spatial post-filters estimated with first-order directional signals. The approach involves applying non-linear transformations on two different spatial post-filters and combining them with weights found by convex optimization of the resulting directivity patterns. The estimation of the post-filters is carried out similarly to the Cross Pattern Coherence (CroPaC) algorithm. The performance of the proposed method is evaluated in a two- and three-speaker scenario with different reverberation times and angular distances of the interfering speaker. The signal-to-interference, signal-to-distortion, and signal-to-artifact ratios are used for evaluation. The results show that the proposed method can improve the spatial selectivity of the post-filter estimated with first-order beampatterns. Using first-order patterns only, it even achieves better spatial separation than the original CroPaC post-filter estimated using first- and second-order signals.