【摘要】 盲分离技术由于不需要知道信号的先验信息而得到广泛应用。利用神经网络信息后向传播的特点,在Infomax算法的基础上,提出一种改进的基于互信息的语音盲分离算法,以神经网络为优化结构,并以输出熵为目标函数,实验证明,算法能很好提取信号的独立分量,完成混合信号的分离。更多还原

【Abstract】 BSS(Blind Source Separation) and ICA(Independent Component Analysis), which do not require prior knowledge of signals, are widely applied. As a generalization of Infomax technique, an improved blind speech signal separation algorithm based on mutual information is presented, which utilizes information back-propagation of neural network. Experimental results demonstrate that, employing optimum structure of neural network and employing output entropy as single objective function, the algorithm is able to extract actual independent components of source, and to separate mixing signals.更多还原