欠通道混叠语音盲分离算法  

Blind Source Separation of Overlapping Speech under Channel-Deficient

作　　者：郭水旺[1] 吕凤英[1] 陶玉福 

机构地区：[1]黄淮学院,河南驻马店463000 [2]河南民权县地方税务局,河南民权476800

出　　处：《河南科学》2010年第4期440-443,共4页Henan Science

基　　金：河南省科技发展计划项目(102102210411)

摘　　要：以超完备基的独立分量分析(Overcomplete ICA)为信道模型,采用自然梯度和最短路径的算法和提出的基于小波变换的Overcomplete ICA算法进行欠通道混叠语音盲分离,两者进行了比较.结果表明,该算法能够成功地分离混叠语音数据.In practical separation of blind source speech, there is a phenomenon： the number of sensors usually less than the number of speech sources,which is refered to ＇separation of blind source speech under channeldeficient＇. To Overcomplete ICA as signal channel model, and propose a Overcomplete ICA algorithm based on wavelet translation which is used in＇separation of blind source speech under channel-deficient＇, and compare with previous natural gradient and shortest path algorithm. Experiments show this algorithm can separate overlapping speech date successfully.

关 键 词：独立分量分析 自然梯度 最短路径 小波变换 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]