快速定点独立向量分析在语音信号盲分离中的应用  

Fast Fixed-point Independent Vector Analysis Algorithms for Speech Blind Separation

作　　者：黄会营[1] 李小光[1] 洪俊芳[2] 

机构地区：[1]洛阳理工学院电气工程与自动化系 [2]中国人民解放军96544部队

出　　处：《河南师范大学学报（自然科学版）》2012年第6期147-149,164,共4页Journal of Henan Normal University(Natural Science Edition)

基　　金：河南省科技厅自然科学基金(201112400450401);洛阳理工学院青年基金(2010QZ16)

摘　　要：针对包括环境噪声的混叠语音信号分离困难问题,提出一种新的频域快速定点独立向量分析模型,对混叠语音信号进行盲分离.该模型中的每个源信号以向量形式混合,模型采用二次泰勒多项式把牛顿迭代算法直接应用于复变量对照函数,通过最优化使牛顿迭代规则更加简单和清晰.对实际采集的语音信号进行仿真,实验证明算法很好地分离了卷积盲源信号,新算法具有良好的收敛性和较高的准确性.In order to solve the difficulty of separating for the aliasing voice signals including ambient noise. A new model, called independent vector analysis, is proposed to blind separation for aliasing speech signal. Each of the source signals are mixed component-wise, Newton iterative algorithm is applied directly to a contrast function of complex variables by using quadratic Taylor polynomials in the model. The Newton derivation rules become more simple and clear by optimization. The actual acquisition of the voice signal is used to simulate based on the new model. The experimental results show that the algorithm has good convergency and high accuracy.

关 键 词：独立向量分析 卷积混合 盲源分离 快速算法 

分 类 号：TN912.3[电子电信—通信与信息系统]