基于EMMD/MMSE低信噪比的语音增强方法  

Research of speech enhancement in low SNR based on EMMD/MMSE

作　　者：张正苏[1] 刘柏森[1,2] 孙美玲[2] 

机构地区：[1]黑龙江工程学院电气与信息工程学院,黑龙江哈尔滨150050 [2]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《黑龙江工程学院学报》2012年第3期40-45,共6页Journal of Heilongjiang Institute of Technology

摘　　要：EMMD分解后的信号虽然解决了EMD中存在的端点效应问题,但是存在音乐噪声,提出一种基于EMMD/MMSE的语音增强方法,该算法利用MMSE解决经过EMMD处理的信号中含有的音乐噪声问题,从而得到很好的语音增强效果。通过实验对增强前后语音信号的信噪比分析表明,基于EMMD/MMSE的语音增强方法在提高语音信号的信噪比、可懂度方面优于传统的增强方法。EMMD can solve the endpoint effect problem exits in EMD, but existing music noise m the sig- nal. Therefore, in this paper proposes a speech enhancement method based on EMMD/MMSE. This algo- rithm makes full use of the characteristics that MMSE can solve the music noise issues exists in EMMD signal processing, so as to obtain good speech enhancement performance. Through doing experiments, comparing the original and enhancement SNR, and subjective judgment, we reach a conclusion that the speech enhancement based on EMMD/MMSE is superior to the traditional enhancement algorithm in im- proving the speech signal SNR and intelligibility.

关 键 词：信号处理 语音增强 极值域均值模式分解 最小均方误差 信噪比 

分 类 号：TN911.7[电子电信—通信与信息系统]