一种改进的双麦克风语音增强算法  

An Improved Dual Microphone Speech Enhancement Algorithm

作　　者：卞金洪[1] 李楠 刘海波 周锋[1] 王如刚[1] BIAN Jinhong;LI Nan;LIU Haibo;ZHOU Feng;WANG Rugang(School of Information Engineering,Yancheng Institute of Technology,Yancheng Jiangsu 224051,China)

机构地区：[1]盐城工学院信息工程学院,江苏盐城224051

出　　处：《电子器件》2022年第6期1448-1452,共5页Chinese Journal of Electron Devices

基　　金：国家自然科学基金项目(61673108);江苏省高等学校自然科学研究重大项目(19KJA110002);江苏省高校自然科学研究面上项目(19KJB510061);江苏省自然科学基金项目(BK20181050);江苏省产学研指导项目(BY2020358,BY2020335)。

摘　　要：提出了一种双通道语音增强算法。该方法结合了最小方差无失真响应波束成形方法和基于超高斯联合的增益函数。同时,采用基于谱通量的语音活动检测来提高波束形成的输出精度。所提出的实时语音增强方法可在助听器等便携式设备上实时运行。最后,利用语音质量和可理解度指标评价了该方法的有效性,并与其他方法进行了比较。客观和主观测试结果表明,该方法在-5 dB、0 dB和+5 dB 3种不同的低信噪比噪声条件下都具有良好的性能。A dual channel speech enhancement algorithm is presented.This method combines the minimum variance distortionless response beamforming method and the super Gaussian joint gain function.At the same time,voice activity detection(VAD)based on spectral flux is used to improve the output accuracy of beamforming.The real-time speech enhancement method proposed can run in real time on portable devices such as hearing aids.Finally,the effectiveness of the proposed method is evaluated with speech quality and intelligibility as the assessment parameters,and compared with other methods.Objective and subjective test results show that the proposed method has good performance under three different noise conditions with low signal-noise ratio(SNR)of-5 dB,0 dB and+5 dB.

关 键 词：助听器 语音增强 MVDR波束成形 超高斯 话音检测 

分 类 号：TN912.35[电子电信—通信与信息系统]