基于拉普拉斯正态混合分布概率密度函数估计的语音分离算法

Speech separation algorithm based on Laplace normal mixture distribution probability density function estimation

由拉普拉斯分布概率密度函数推导得出的符号函数作为语音信号分离算法中的激活函数,目前广泛应用在基于独立分量分析的语音分离算法之中。为了提高语音分离算法的收敛速度以及分离性能,提出把拉普拉斯正态混合分布概率密度函数作为语音信号概率密度函数的估计,得到一个更加适合语音信号分离的激活函数,基于此函数提出一种快速语音分离算法。把推导出的激活函数应用于独立分量分析(ICA)的自然梯度算法中进行计算机仿真实验,可验证本文算法的收敛性能和分离性能。

Speech separation algorithms using sign function deduced by Laplace distribution function as the activation function is used universal.To achieve fast convergence and good performance in real time speech separation,a fast speech separation algorithm was proposed in this paper.Taking Laplace normal mixture distribution function as a statistical estimation model for speech signals,a nonlinear activation function which is more suitable for speech separation is obtained.And on the basis of this function,using this nonlinear activation function in natural gradient based ICA algorithm in computer simulations,we get the results which prove the above mentioned properties.