动态阈值谱法语音增强  

A METHOD OF DYNAMIC THRESHOLD SPECTRUM FOR SPEECH ENHANCEMENT

作　　者：陆生礼[1] 余崇智[1] 魏荣爵[1] 

机构地区：[1]南京大学声学研究所

出　　处：《南京大学学报（自然科学版）》1996年第2期218-223,共6页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金

摘　　要：根据人耳能从噪声中提取有用信息的听觉特征，并结合语音信号的基本特征，提出并研究了一个适合于语音增强的听觉模型；实验结果表明，这个方法不仅在提高语音信噪比方面，而且在减小语音失真度方面均有较好的改善。This paper studies and simulates the hearing evidence that human auditory system can select some useful information from noisy disturbance. Based on some hearing evidence, we find the relation between hearing response and hearing threshold. Human auditory system adjusts threshold value of hearing response according to the useful sound information that we want to obtain. It is assummed that the auditory system fetched and recognized speech from background noise or disturbace just because of the spectrum of the dynamic threshold value. Thus an auditory model of speech enhancement is presented. Base on this model, some experiments about niosy speech are presented. The results show that this method of speech enhancement yields lower distortion as well as higher signal to noise (S/N) ratio than conventional methods.

关 键 词：听觉模拟 语音增强 语音通信 动态阈值谱法 

分 类 号：TN912.3[电子电信—通信与信息系统]