毕业设计（论文）-基于ICA算法的混合语音信号分离精选.doc

（ 二〇一二 年 六 月 摘 要 现阶段对语音识别、医学、生物、通信等领域中的盲源混合语音信号进行分离有着较高的要求。在任意环境下，会普遍遇到从多维信号源提取或分离出某一独立语音信号这种类似于“鸡尾酒会”的问题。这时，就迫切需要一种合理可靠的方法将观测信号分解为若干独立份。 由于信号源是不可观测的，并不知道信号源是由哪些独立信号混合而成。正是，由于ICA算法建立在独立统计之上，适用于解决这种分离混合语音的问题。本文参考算法中独立统计的原则。建立信号源的目标函数，对目标函数进行预处理，运用负熵极大化法求出原独立信号向量的预估向量Y。ICA算法中的负熵极大化法具有使分离后的语音快速发散、信息极大化的特点。所以本文选择ICA算法中的负熵极大化法来分离混合语音。这就是使用ICA算法独立统计原则推断混合系统的独立源信号的基本原因。 经过对实验仿真图与原独立语音图比较，实验得出如下结论：利用ICA算法处理盲源混合信号，分离后得到的独立语音清晰、准确、算法优化、方便快捷。 关键词：独立分量分析；ICA算法；熵极大化法 Abstract At the present stage speech recognition at medicine, biology, communications and other fields for BSS（Blind Source Speech Separation）has an important position. Whatever the environment is，we need deal with something about separate independent voice from the unknow mixed voice. This situation is nearly “Cocktail party”.So,we need find a good way to solve this problem. As source is not observed, we don`t know the source be made up by which independent signals . ICA algorithm is apply to solve the problem which mixed voice of this separation. This reference algorithm independent statistical principles to establish the source of the objective function. Pretreatment of the objective function, the use of negative entropy maximization method to calculate the estimated vector Y of the original independent signal vector. ICA algorithm in a negative entropy maximization method after the separation of voice rapid divergence, the characteristics of information maximization. Therefore negative entropy maximization method in the ICA algorithm to separate the mixed voice. This is the basic reason of the independent source signals using the ICA algorithm independent statistical principles to infer hybrid systems. Experimental results show that：using the algorithm of ICA (Independent Component Analysis) process the blind source mixed-signal.We can easy get an independent signal which is very clear, accurate, algorithm optimization and convenient. Keywords：Independent component analysis；Algorithm of