一种基于小波变换的独立语音量分离方法  

An Approach to Independent Speech Components Separation Based on Wavelet Transform

作　　者：任济生[1] 任鹏[2] 吉爱国[2] 曹茂永[3] 

机构地区：[1]山东科技大学机械与电子工程学院,山东青岛266510 [2]青岛理工大学通信与电子工程学院,山东青岛266033 [3]山东科技大学信息与电气工程学院,山东青岛266510

出　　处：《电子器件》2008年第3期1019-1022,共4页Chinese Journal of Electron Devices

基　　金：国家自然科学基金资助项目(60472047);山东省优秀中青年科学家科研奖励基金资助项目(02BS011)

摘　　要：提出了一种基于小波变换的独立语音量分离方法,以改进卷积叠加语音分离方法。该方法在小波域进行独立元分析,把高阶去相关矩阵分解为小波子空间中一系列相应的低阶去相关矩阵,并在保证语音短时平稳性的基础上通过低阶处理跟踪时变的室内(存在室内回响)传输函数。实验结果表明,提出的方法改善了传统独立元分析在室内回响环境中独立语音量分离的效果。A separation method for independent speech components based on wavelet transform is proposed to improve the analysis of convulsive mixed speech. Separation of independent speech components is implemented in wavelet domain where Independent Component Analysis is applied. Low order processing is used to trace the variation of room echo-existed transmission function by analyzing high order decorrelated matrix into a series of lower order ones in wavelet subspaces so that the short-time stationality of speech is guaranteed. Experimental results indicate that the proposed method improves the performance separation of independent speech components by Independent Component Analysis in a room-echo existed environment.

关 键 词：语音量分离 小波变换 独立元分析 室内回响环境 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]