摘要
提出用小波变换和两步自适应盲分离算法相结合的方法来进行语音分离。首先,利用小波变换分别对各含噪混叠语音进行消噪;然后,利用代价函数的极值点特性分别获得混合信号和白化信号的特征向量矩阵,实现自适应盲分离过程;最后,进一步对分离信号进行矢量归一和再消噪处理,得到各个语音源信号的最终估计。实验结果表明此方法取得了很好的分离效果。
A new adaptive blind source separation algorithm is presented based on wavelet transform and second order statistics of the signal. Firstly, the noisy mixtures with discrete wavelet transform are denoised. Secondly, the eigenvector matrices of the mixed signals and the whitened signals are obtained using the point property of a special cost function. Finally, the separated signals are got. The simulation shows that the algorithm can get good vector normalized, then the estimated source speech are separation performance.
出处
《电声技术》
2008年第5期54-56,68,共4页
Audio Engineering
关键词
语音盲分离
小波变换
白化处理
噪声消除
代价函数
speech blind separation
wavelet transform
whitening process
denoising
cost function
