基于多窗谱估计与NLMS自适应滤波算法的单通道语音增强  

Single-channel Speech Enhancement Based on Multi-window Spectral Estimation and NLMS Adaptive Filtering Algorithm

作　　者：孙逸飞 涂振宇[1] 相敏月 马飞 方强 SUN Yifei;TU Zhenyu;XIANG Minyue;MA Fei;FANG Qiang(School of Information Engineering,Nanchang Institute of Technology,Nanchang Jiangxi 330099,China)

机构地区：[1]南昌工程学院信息工程学院,江西南昌330099

出　　处：《通信技术》2022年第11期1419-1427,共9页Communications Technology

基　　金：江西省水利厅科技项目(KT201639);江西省科技厅重点研发计划(20151BBE50077)。

摘　　要：多数传统语音增强算法是基于平稳噪声下分析的,且没有从语音质量及可懂度角度全面衡量增强性能。因此,提出了基于多窗谱估计与归一化最小均方(Normalized Least Mean Square,NLMS)自适应滤波算法的单通道语音增强方案。首先利用多窗谱估计谱减法(Multiwindow Spectral Subtraction,MSS)解决谱减法产生的“音乐噪声”问题;其次将估计出的期望信号与纯净参考信号的差值作为误差信号,由自适应滤波的NLMS算法代替传统的最小均方(Least Mean Square,LMS)算法,以降低滤波器成本及运算时间,求取滤波器权系数值,并不断迭代更新修正滤波器;最后分析了所提算法在不同噪声环境下的增强性能,并与传统的各种谱减算法对比,从语音质量及可懂度出发衡量语音增强效果。结果表明,所提算法的增强效果优于各类谱减法。Most traditional speech enhancement algorithms are analyzed based on stationary noise, and the enhanced performance is not comprehensively measured from the perspective of voice quality and intelligibility. Therefore, a single-channel speech enhancement scheme based on multi-window spectral estimation and NLMS adaptive filtering algorithm is proposed. First, the MSS(Multi-window Spectral Subtraction)is used to solve the problem of “music noise” caused by spectral subtraction. Then, the difference between the estimated expected signal and the pure reference signal is taken as the error signal, and the NLMS(Normalized Least Mean Square) algorithm of adaptive filtering replaces the traditional LMS(Least Mean Square) algorithm to reduce the filter cost and operation time, to obtain the filter weight coefficient value, and to constantly update and modify the filter iteratively. Finally, the enhancement performance of the proposed algorithm under different noise environments is analyzed, and compared with various traditional spectral subtraction algorithms to measure the speech enhancement effect from the perspective of speech quality and intelligibility. The results indicate that the enhancement effect of the proposed algorithm is better than that of various spectral subtraction algorithms.

关 键 词：语音增强 谱减法 多窗谱 自适应 

分 类 号：TN912[电子电信—通信与信息系统]