摘要:
语音信号盲源分离是在不知道源信号和传输信道任何先验知识的情况下,仅根据输入语音源信号的统计特性,通过观察信号恢复出各个独立源信号的过程.基于负熵的FastICA算法的最大优点在于其收敛速度较快,是一种性能较好的学习算法,在语音信号盲源分离应用中具有较高的实用价值,但该算法存在计算量大的缺点.针对这一问题,提出了以负熵最大化作为目标函数,采用牛顿迭代法求解FastICA算法的方法.但由于牛顿迭代法求解非线性方程计算量较大,需改进雅克比矩阵算式,并使用相关系数来观察分离效果的优劣.仿真分析结果证明,改进后的算法由于未增加雅克比矩阵运算次数,所以收敛的迭代次数大幅度减少,使程序运行时间显著减小.该算法在兼顾算法收敛速度的同时,提高了算法的性能,较好地恢复了源语音信号.%Voice signal blind source separation is the recovery process of each independent source signal,by observing signals and only in accordance with the statistical characteristics of the input speech source signals, without any prior knowledge about the source signal and transmission channel. The negative entropy based FastICA algorithm is a learning algorithm with better performance,its biggest advantage is fast convergence, and it has a good practical value in the application of blind source separation of voice signals;but the disadvantage of the algorithm is large amount of calculation. To solve this problem,the negative entropy maximization is used as the objective function,and FastICA formula is solved by Newton iterative method. Because the process of Newton iterative method for solving nonlinear equations still has a large amount of calculation, the Jacobian matrix formula is studied and improved;the correlation coefficient is used to observe the effect of separation. The results of simulation analysis prove that the improved algorithm does not increase the operation times of Jacobian matrix,so the number of iterations of convergence is greatly reduced and the running time of programs significantly decreased. Besides convergence speed,the algorithm also improves the performance,it well recovers the source voice signal.