论文部分内容阅读
针对语音信号的卷积混迭模型,利用不同语音信号之间的近似独立和短时平稳特性,提出一种基于信号二阶统计量的联合块对角化方法,解决超定卷积盲分离问题。该方法采用非对角线上各子矩阵F-范数的平方和作为联合块对角化性能的评判准则,将原四次代价函数转化为一组较为简单的二次子代价函数,每一子代价函数用于估计酉混迭矩阵的一个子矩阵。依次最小化各子函数,迭代搜索代价函数最小点,得到混迭矩阵的估计。理论分析及实验结果表明,所提方法不仅能够达到与类Jacobi经典方法同样好的分离效果,并且具有更低的计算复杂度、更快的收敛速度和对传输信道阶数、迭代初始值不敏感的特点。
Aiming at the convolution and aliasing model of speech signal, a joint block diagonalization method based on signal second-order statistics is proposed to solve the problem of overdetermined convolutional blind separation by using the approximate independent and short-term stationary characteristics of different speech signals. . In this method, the square sum of the F-norm of each sub-matrix on the non-diagonal is used as the criterion for the diagonalization of the joint block. The original four-cost function is transformed into a simple quadratic sub-cost function, Sub-cost function is used to estimate a submatrix of the unitary mixing matrix. In turn, each sub-function is minimized to iteratively search for the minimum point of the cost function to obtain the estimation of the aliasing matrix. The theoretical analysis and experimental results show that the proposed method not only achieves the same good separation effect as the Jacobi-like classical method, but also has lower computational complexity, faster convergence rate and insensitivity to the transmission channel order and iterative initial value specialty.