基于小波变换多分辨率特性的语音消噪方法的研究  

Research of speech de-noising method based on multi-resolution of wavelet transform

作　　者：王惠琴[1] 张秋余[1] 薛建斌[1] 

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《计算机工程与设计》2006年第2期235-237,296,共4页Computer Engineering and Design

基　　金：甘肃省自然科学基金项目(ZS022-A25-032)

摘　　要：小波分析是一种信号的时间——尺度分析方法,特别适合于非平稳信号的分析,具有多分辨率分析的特性,而且在时频两域都具有表征信号局部特征的能力。通过分析语音信号的特性,利用小波变换的多分辨率分析特性,提出了首先对信号进行清浊音判断,其次运用多尺度多阈值方法来抑制包含有噪声的语音信号在不同尺度上的噪声小波系数,从而实现在重构语音信号中消噪的目的,并通过计算机仿真结果验证了该方法的有效性。Wavelet transform is an analytical method of the time-frequency, which is especially suitable for the analysis of non-stable signal. The method has a character ofmultiresoulution analysis, and has an ability of showing singars part character in time-frequency domain. Firstly, a judjement between voiceless sounds and voiced sounds for signal was proposed by the character of multiresoulution analysis of wavelet, and then the aim of reconstructed speech de-noising was realized by using an method that wavelet transform coeffici- ents of noise-speech signal were filtered by changing threshold on the different scale. The computer de-noising emulated results are given to show that this method is very effective.

关 键 词：小波变换 多分辨率分析 语音信号 消噪 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]