首页 | 本学科首页   官方微博 | 高级检索  
     

基于改进EMD的麦克风阵列语音增强研究
引用本文:李恝,吴海彬,叶锦华. 基于改进EMD的麦克风阵列语音增强研究[J]. 福州大学学报(自然科学版), 2019, 47(6)
作者姓名:李恝  吴海彬  叶锦华
作者单位:福州大学 机械工程及自动化学院 福州 350000,福州大学 机械工程及自动化学院 福州 350000,福州大学 机械工程及自动化学院 福州 350000
基金项目:国家自然科学基金(51605093)
摘    要:提出一种将改进EMD与麦克风阵列MVDR自适应波束形成相结合的语音增强方法。该方法利用互相关系数阈值法去除将EMD算法分解后的的虚假IMF分量,结合各阶IMF分量的自相关函数特性准确获取信号与噪声的主导IMF分量分界点,然后对所有噪声主导的IMF分量进行小波阈值去噪,接着将所有剩余IMF分量进行MVDR波束形成获得增强语音信号。改进EMD算法避免了在高信噪比条件下的信号失真,与MVDR波束形成相结合,满足了MVDR窄带特性要求,增强了麦克风阵列抗干扰能力。实验结果证明了方法的有效性。

关 键 词:语音增强  经验模态分解  麦克风阵列  MVDR  
收稿时间:2019-02-25
修稿时间:2019-03-29

Microphone array speech enhancement research based on improved EMD.
LI Ji,WU Haibin and YE Jinhua. Microphone array speech enhancement research based on improved EMD.[J]. Journal of Fuzhou University(Natural Science Edition), 2019, 47(6)
Authors:LI Ji  WU Haibin  YE Jinhua
Affiliation:School of Mechatronics Engineering and Automation,Fuzhou University,School of Mechatronics Engineering and Automation,Fuzhou University,School of Mechatronics Engineering and Automation,Fuzhou University
Abstract:A speech enhancement method combining improved EMD with microphone array MVDR adaptive beamforming is proposed. In this method, the false IMF components decomposed by EMD algorithm are removed by cross-correlation coefficient threshold method, and the demarcation points of dominant IMF components of signal and noise are obtained accurately by combining the autocorrelation function characteristics of each order of IMF components. Then, the wavelet threshold denoising is applied to all noise-dominant IMF components, and then all residual IMF components are beamformed by MVDR to obtain enhanced speech signals. The improved EMD algorithm avoids the signal distortion under the condition of high SNR, combines with MVDR beamforming, meets the narrowband characteristics of MVDR, and enhances the anti-jamming ability of microphone array. The experimental results show the effectiveness of the method.
Keywords:speech enhancement   EMD   microphone array   MVDR
点击此处可从《福州大学学报(自然科学版)》浏览原始摘要信息
点击此处可从《福州大学学报(自然科学版)》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号