基于维纳滤波与理想二值掩蔽的数字助听器语音增强算法  

Speech Enhancement Based on Wiener Filteringin Digital Hearing Aids

作　　者：戴红霞[1] 唐於烽 赵力[2] DAI Hongxia;TANG Yufeng;ZHAO Li(Department of Electronics Communications Engineering,Jiangsu Information Occupation Technical College,Wuxi Jiangsu 214061,China;School of Information Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]江苏信息职业技术学院电子信息工程系,江苏无锡214153 [2]东南大学信息科学与工程学院,南京210096

出　　处：《电子器件》2019年第4期994-997,共4页Chinese Journal of Electron Devices

基　　金：国家自然科学基金项目(61673108);江苏高校品牌专业建设工程项目(2017116)

摘　　要：研究了基于先验信噪比估计的维纳滤波增强算法。针对基于先验信噪比估计的维纳滤波法未能有效提高语音可懂度的缺点进行了研究与改进:首先,引入了以MMSE为准则的两步先验信噪比估计取代原“直接判决”法;其次,针对信噪比为负的区域,增益函数误差的过高估计会严重降低语音可懂度的问题对增益函数进行了改进,放大了负信噪比区域的噪声谱,从而降低增益函数被过高估计的可能性;最后,结合理想二值掩蔽的技术,对处理后的语音信号进行后处理。实验结果表明,改进的方法能够有效的提高语音可懂度并降低语音信号的畸变。This paper focuses on the speech enhancement based on a priori SNR estimation Wiener Filtering.The methodbased on priori SNR estimation Wiener Filtering cannot effectively improve the speech intelligibility.Some research is done to solve this disadvantage.Firstly,this paper uses the MMSE-based two-step a priori SNR estimate approach to replace the“decision directed”method.Secondly,the gain function is modified because of the problem that the gain function overestimation errors in frequency bins with negative SNR are particularly harmful to speech intelligibility.Finally,Combined with the ideal binary masking technique,the processed speech signal is post-processed.The comparative experiments with the original method demonstrate the effectiveness of the improved method and the reduction of distortion of speech signal.

关 键 词：数字助听器 维纳滤波 先验信噪比 谐波重构 语音可懂度 

分 类 号：TN912.34[电子电信—通信与信息系统]