基于人类听觉系统的多频带语音增强  

Multi-band spectral subtraction method for speech enhancement based on human auditory system

作　　者：沈圆圆[1] 顾济华[1] 陶智[1,2] 赵鹤鸣[2] 谈雪丹[1] 

机构地区：[1]苏州大学物理科学与技术学院,江苏苏州215006 [2]苏州大学电子信息学院,江苏苏州215006

出　　处：《计算机工程与应用》2011年第21期145-148,共4页Computer Engineering and Applications

基　　金：国家自然科学基金No.60572076;江苏省高校自然科学基金(No.09KJD510005)~~

摘　　要：提出了一种基于人类听觉系统的多频带非线性谱减法来进行语音增强。根据人耳听觉特性,将含噪语音信号分在24个临界频带内,由各频带的不同信噪比来确定对应的谱减参数值。实验结果证明,在相同实验条件下,与功率谱减法(PSS)、非线性谱减法(NSS)和传统多频带谱减法(MBSS)相比,该方法增强后的语音信号具有更高的输出信噪比;能更好地消除背景噪声,抑制残留噪声;增强后的语音具有更好的可懂度和清晰度。A multi-band nonlinear spectral subtraction method is proposed with the demand on improving the quality of speech signal corrupted by colored noise.According to the human auditory system,the noisy speech signals are nonlinearly divided into 24 non-overlapping critical bands.The amount of noise needed to be subtracted is determined by different Signal-to-Noise Ratio（SNR） of each band.Under the same experimental conditions,test results show that compared to the Power Spectral Subtraction method（PSS）,Nonlinear Spectral Subtraction method（NSS） and the Multi-Band Spectral Subtraction（MBSS）,the proposed method can reduce the residual noise and background noise more effectively.The enhanced speech signals of the proposed algorithm have better clearness and intelligibility besides higher output SNR.

关 键 词：听觉系统 多频带 谱减法 语音增强 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]