基于Borel测度峰值判定的欠定混合盲语音信号分离

Blind speech signals separation based on Borel measure peaks for under-determined mixtures

简要介绍稳定分布的特征函数及其Borel测度表示,给出了Borel测度的估计方法,并利用Borel测度的峰值确定混合矩阵的基矢量,从而可以确定各个独立分量,实现信号的盲分离。计算机模拟和分析表明,这种算法是一种在高斯和分数低阶Alpha稳定分布噪声条件下具有良好韧性的独立分量分析与盲源分离方法,在盲语音混合信号的分离应用中也得到了很好的效果。

The statistical characteristics of stable distribution was briefly introduced, and a new method for identifying the independent components of an Alpha-stable random vector for under-determined mixtures was proposed. The method was based on an estimate of the discrete spectral measure for the characteristic function of an Alpha-stable random vector. Simulations demonstrate that the proposed method can identify independent components and the basis vectors of mixing matrix in the so-calied under-determined case of more sources than mixtures, and obtained the good effect in the speech signals separation application.