结合语音包络的变步长LMS语音自动增益控制  

Variable Step Size LMS Speech Automatic Gain Control by Combining Speech Envelope

作　　者：蒋学仕 JIANG Xueshi(Southwest China Institute of Electronic Technology,Chengdu 610036,China)

机构地区：[1]中国西南电子技术研究所,成都610036

出　　处：《电讯技术》2022年第2期252-258,共7页Telecommunication Engineering

摘　　要：针对现有最小均方误差(Least Mean Square,LMS)语音自动增益控制(Automatic Gain Control,AGC)算法其功率参量不能准确表示语音振幅、增益控制效率较低以及增益处理后语音失真较大等问题,提出使用语音包络作为参量的变步长LMS-AGC算法。首先按帧计算语音包络作为参量,提升参量表示语音振幅的准确度;然后利用递归平滑的语音包络误差平方,调节LMS算法的步长,提升算法的增益响应速度;最后将算法与低失真的限幅处理机制结合,解决现有LMS-AGC算法增益处理后存在过门限尖峰的问题。实验结果表明,相比现有基于功率的变步长LMS-AGC算法,基于语音包络的变步长LMS-AGC算法的增益响应速度更快,增益处理后客观语音质量评估(Perceptual Evaluation of Speech Quality,PESQ)平均提升0.4分,并且迭代运算量可以下降一个数量级。To solve the problem that the power parameter of the existing least mean square(LMS)automatic gain control(AGC)algorithm can not accurately represent the speech amplitude,the gain control is inefficient and the speech distortion is large,a variable step LMS-AGC algorithm using speech envelope as parameter is proposed.First,the speech envelope is calculated as a parameter by frame to improve the accuracy of the speech amplitude.Then,the recursively smoothed error square of speech envelope is used to adjust the step size of LMS algorithm and improve the gain response speed.Finally,the algorithm is combined with the low-distortion amplitude limiting to avoid the problem that the LMS-AGC algorithm has over-threshold spike.Experimental results indicate that the algorithm based on speech envelope has faster gain response speed,and the perceptual evaluation of speech quality(PESQ)is improved by 0.4 points on average,and the iterative computation can be reduced by an order of magnitude compared with the existing power-based variable step size LMS-AGC algorithm.

关 键 词：语音自动增益控制 语音包络 变步长 低失真 

分 类 号：TN912.34[电子电信—通信与信息系统]