摘要
针对低信噪比下语音增强困难的问题及经验模态分解(empirical mode decomposition,EMD)在语音增强中的有效应用,提出了一种基于EMD的前后置滤波语音增强算法.将含噪语音信号进行卡尔曼滤波后做EMD分解,采用能量阈值法判断出含有残余噪声的本征模态函数(IMF),对其小波变换后与其余IMF重构为最终的语音增强信号.实验表明,在输入信噪比为-10~5dB的不同噪声环境下,以时域分段信噪比及语音质量感知评测PESQ作为评价指标,效果均优于单独的EMD、小波软阈值法及卡尔曼滤波算法,是一种有效的语音增强算法.
To solve the problem of difficulty in speech enhancement under low SNR conditions,this paper proposes a pre-and post-filtered speech enhancement algorithm based on empirical mode decomposition.The algorithm first performs Kalman filter on the speech signal containing noise,and then performs empirical mode decomposition.And uses energy threshold algorithm to determine the intrinsic modal function with residual noise.After the wavelet transform,the intrinsic modal function is reconstructed with the rest of the intrinsic modal function to become the final speech enhancement signal.In the experiment,SNR of the noise environment was set within the range of-10dB to 5dB,and the time domain segmented signal-to-noise ratio and speech quality perception evaluation were used as evaluation indices.Experimental results show that the effect of speech enhancement is better than that using EMD,wavelet soft thresholding,and Kalman filtering algorithm respectively.
作者
潘晴
冉福星
李雅昆
Pan Qing;Ran Fuxing;Li Yakun(School of Information Engineering,Guangdong University of Technology,Guangzhou 510006,China)
出处
《河南师范大学学报（自然科学版）》
CAS
北大核心
2018年第3期33-39,共7页
Journal of Henan Normal University(Natural Science Edition)
基金
国家自然科学基金(61372173)
关键词
卡尔曼滤波
经验模态分解
能量阈值
小波变换
Kalman filter
empirical mode decomposition
energy threshold
wavelet transform
