中文摘要：

利用带噪语音经特征基函数矩阵转换后所具有的稀疏特性,用最大似然估计方法对转换后得到的稀疏分量进行非线性压缩去噪,然后再经过反变换和重构恢复出原始语音信号的估计。特征基函数矩阵反映了语音数据本身的统计特性。仿真结果表明,与小波消噪法相比利用稀疏编码方法能极大程度地抑制背景噪声。

英文摘要：

The sparse coding speech enhancement utilizes the sparsity of the components,which are obtained by transforming the noisy data through the basis functions matrix.Using the maximum likelihood estimation（MLE）,the sparse components can be nonlinearly de-noised,then after the inverse transformation and reconstruction,the estimated original speech was obtained.The basis functions matrix reflects the statistical characteristics of speech data itself,so this proposed method is of great rationality.Computer simulated results show that the sparse coding approach present excellent speech enhancement effect compared with wavelet de-noising in inhibition backgrounds-noisy. 【Keyword】：