基于最大信息熵的小波包阈值去噪语音增强算法  

A Wavelet Packet Threshold De-Noising Algorithm for Speech Enhancement Based on Maximum Information Entropy

作　　者：杨桂芹[1] 徐红莉[1] 

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070

出　　处：《测控技术》2011年第10期12-14,17,共4页Measurement & Control Technology

基　　金：甘肃省自然科学基金资助项目(1010RJZA065)

摘　　要：去噪算法在语音增强中占有极为重要的地位。而传统的小波阈值去噪算法会不可避免地造成部分有用语音信号的损失。为了更好地对含噪语音信号进行去噪,选用小波包分析法进行语音分解,采用一种新的阈值函数,同时基于最大信息熵的原理确定了阈值和加权阈值函数中的权因子。仿真结果表明,该算法能够很好地抑制噪声,保证重建信号质量。无论在听觉效果上还是信噪比指标上,均具有更好的去噪性能。De-noising algorithm plays a very important position in the speech enhancement, while the traditional wavelet threshold de-noising algorithm will cause loss of part of useful speech signals inevitably. In order to de- noise the speech signals with noise better, a wavelet packet analysis method for voice decomposition is seleted, and a new threshold function is used. The threshold and the weighted factors in the weighted threshold function are determined by the theory of maximum information entropy. The simulation results show that the posed algorithm can suppress noise better, and ensure the quality of the reconstructed signal. The de-noising effect is better not only in the audio effects but also in the signal to noise ratio.

关 键 词：最大信息熵 阈值 阈值函数 小波阈值去噪 语音增强 

分 类 号：TN912[电子电信—通信与信息系统]