Incorporating the Noise Statistics in Acoustic Multi-channel Equalization

Acoustic multi-channel equalization techniques, such as the regularized partial multi-channel equalization technique based on the multiple-input/output inverse theorem (RP-MINT), are able to achieve a high dereverberation performance in the presence of room impulse response perturbations but may lead to additive noise amplification. This paper proposes to directly extend the RP-MINT technique by incorporating the noise statistics in the reshaping filter design, such that joint dereverberation and noise reduction is achieved. In addition to the regularization parameter used in the RP-MINT technique, a weighting parameter is introduced to trade off between dereverberation and noise reduction. To automatically determine the regularization and weighting parameters, a novel non-intrusive procedure based on the L-hypersurface is proposed. Simulation results using instrumental performance measures show that the proposed technique maintains the high dereverberation performance of the RP-MINT technique, while improving the noise reduction performance.