基于自适应逼近残差的稀疏表示语音降噪方法  

Adapted Stopping Residue Error Based Sparse Representation for Speech Denoising

作　　者：周伟力[1] 贺前华[1] 王亚楼[1] 庞文丰 ZHOU Weili HE Qianhua WANG Yalou PANG Wenfeng(School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640, China)

机构地区：[1]华南理工大学电子与信息学院,广州510640

出　　处：《电子与信息学报》2017年第2期309-315,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(61571192);广东省公益项目(2015A010103003)~~

摘　　要：该文提出一种基于自适应逼近残差的稀疏表示语音降噪方法。在字典学习阶段基于K奇异值分解(K-Singular Value Decomposition,K-SVD)算法获得干净语音谱的过完备字典,在稀疏表示阶段基于权重因子调整后的噪声谱和估计的交叉项对逼近残差持续自适应地更新,并采用正交匹配追踪(Orthogonal Matching Pursuit,OMP)方法对干净语音谱进行稀疏重构。最后结合估计的干净语音谱与带噪语音相位,通过傅里叶逆变换获得重构的干净语音。实验结果表明所提方法在不同噪声和信噪比条件下相比标准的谱减法,稀疏表示语音降噪算法和基于自回归隐马尔可夫模型的降噪方法有更好的降噪效果。A sparse representation speech denoising method based on adapted stopping residue error is proposed. Firstly, an over complete dictionary of the clean speech power spectrum is learned by the K-Singular Value Decomposition （K-SVD） algorithm. In the sparse representation stage, the stopping residue error is adaptively achieved according to the estimated cross terms and the noise spectrum which is adjusted by a weighted factor, and the Orthogonal Matching Pursuit （OMP） approach is applied to reconstruct the clean speech spectrum from the noisy speech. Finally, the clean speech is re-synthesis via the inverse Fourier transform with the reconstructed speech spectrum and the noisy speech phase. The experiment results show that the proposed method outperforms the standard spectral subtraction, sparse representation based speech denoising algorithm and the AutoRegressive Hidden Markov Model （AR-HMM） based speech denoising method in terms of subjective and objective measure.

关 键 词：语音降噪 稀疏表示 K奇异值分解 正交匹配追踪 

分 类 号：TN912.3[电子电信—通信与信息系统]