Speaker
Description
Adaptive beamforming in the spherical harmonics (SH) domain is widely used for spatial filtering and is often formulated as a minimum variance optimization, which suppresses interfering components based on directional energy. Perceptual disturbance, however, is not determined by energy alone but also depends on spectral and temporal characteristics, leading to situations in which energetically weak yet perceptually salient components remain insufficiently attenuated. This work proposes an annoyance-weighted beamforming formulation that incorporates psychoacoustic metrics directly into the beamformer design. Directional signals are obtained via plane-wave decomposition and evaluated using Di’s psychoacoustic model, combining loudness, sharpness, roughness, fluctuation strength, and tonality into a single annoyance measure. These directional estimates are then used to construct an annoyance-weighted spatial covariance matrix, replacing the conventional covariance in the minimum variance distortionless response (MVDR) formulation while retaining its closed-form solution. A power-weighted reference formulation is introduced for comparison between perceptual and energy-based spatial filtering alongside the baseline MVDR. The method was evaluated using simulated acoustic scenes with varying interferer characteristics and source distributions based on the spherical harmonics formulation. The results demonstrate clear differences between the spatial distributions of annoyance and energy. The proposed method yields distinct beamforming patterns and achieves improved selectivity of perceptually annoying interference relative to energy-based approaches. A complementary trade-off is observed with conventional MVDR beamforming, which more effectively reduces overall residual energy.