改进的信息最大化语音盲分离算法

提出一种改进的基于信息最大化的语音盲分离算法,克服了以往算法收敛速度慢,串音误差大的缺点.新算法分析了多种非线性函数的转换特性,采用一种新的更合适的非线性函数,并依此推导出分离算法的学习规则.实验表明,改进的算法有效实现了混叠语音信号的盲源分离,收敛速度更快,串音误差更小,取得了良好、稳定的分离效果.

Improved Speech Blind Separation Algorithm Based on Information Maximum

An improved speech blind separation algorithm based on information maximum is proposed to overcome the shortcomings of slow convergence speed and high crosstalk error of the past algorithms.New algorithm uses a new,more suitable non-linear function by analyzing the conversion feature of a variety of non-linear functions,and deduces the learning rule of the separation algorithm according to it.Experiments show that the improved algorithm can effectively realize blind source separation of speech signals mixture,and the separation achieves good,stable results with faster convergence speed and smaller crosstalk error.