摘要
由于小波阈值在语音去噪中阈值的单一性,本文提出了一种基于稀疏表示理论的新的去噪算法.该算法首先用K-SVD字典学习得到信号在字典下的稀疏表示,其次用形态成分分析(Morphological Component Analysis,MCA)将语音信号分为高幅部分和低幅部分,最后用重构方法对各部分语音信号进行重构及合成.通过实验仿真,并与小波阈值去噪方法比较,本文所提方法的去噪效果更好,鲁棒性更强.
Due to the wavelet threshold＇s singleness in speech de-noising,this paper presents a novel speech de-noising approach based on sparse representations （SR）. Firstly,the sparse represen- tation of the noisy signal is achieved through K-SVD dictionary learning algorithm. Secondly,Mor- phological Component Analysis（MCA） method is used to separate the speech signals into high-am- plitude part and low-amplitude part. Finally, the two parts of the signals are reconstructed and put together. The simulation result shows that the method has better deoising effect and stronger ro- bustness compared to the wavelet threshold approach.
出处
《北方工业大学学报》
2013年第3期6-11,共6页
Journal of North China University of Technology
基金
国家自然科学基金资助项目(No.61170327)
关键词
稀疏表示
K-SVD
语音去噪
形态成分分析
sparse representation
K-SVD
speech de-noising
morphological component analysis
