A Time-Domain Blind Convolutive Separation Algorithm for Speech Signals Based on Second-Order Statistic
基于二阶统计量的语音信号时域卷积盲分离算法

For blind convolutive separation of speech signals, a cost function improved from the least squares fitting function is developed based on second-order statistic, which simplifies the quartic-function with respect to the mixture matrix into three quadratic functions. The iternative approach with three sub-steps is proposed to perform the non-unitary joint block-diagonalization. In each sub-step, a closed solution is derived by minimizing the cost function associated with one parameter-group while fixing the others. Furthermore, the feature of low computational complexity is analytically proven. Compared with ZJBD, the simulations illustrate that the proposed algorithm has the merits of robust initialization selection and better estimate accuracy.