Deficient-basis-complementary rank-constrained spatial covariance matrix estimation based on multivariate generalized Gaussian distribution for blind speech extraction

Abstract Rank-constrained spatial covariance matrix estimation (RCSCME) is a blind speech extraction method utilized under the condition that one-directional target and diffuse background noise are mixed. In this paper, we propose new model extension of RCSCME. RCSCME simultaneously conducts both deficient rank-1 component complementation matrix, which incompletely estimated by preprocessing methods such as independent low-rank analysis, source parameters. conventional RCSCME, between two parameters constituting component, only scale estimated, whereas other parameter, basis, fixed in advance; however, how to choose basis not unique. proposed model, also regard parameter estimate. As generative an observed signal, super-Gaussian generalized Gaussian distribution, achieves better separation performance than distribution Assuming derive majorization-minimization (MM)- majorization-equalization (ME)-algorithm-based update rules for basis. particular, among innumerable ME-algorithm-based rules, successfully find rule with mathematical proof supporting fact step larger MM-algorithm-based rule. We confirm outperforms several simulated conditions real condition.