基于遗传神经网络的语音信号盲分离算法

通过分析基于神经网络的经典盲分离算法具有容易陷入局部极小点,从而导致收敛速度慢和分离效果不准确的缺点,本文首先利用遗传神经网络算法对分离权值进行初始化,然后通过选择操作、交叉操作和变异操作,进行样本训练控制,在整个搜索空间进行搜索,得到分离矩阵最优值,最后实现了语音信号的盲分离。实验表明：该算法具有分离速度快、效果明显等特点。

Blind separation algorithm for audio signal based on genetic algorithm and neural network

The blind separation of audio signal is an important application of blind signal separation technology. The traditional separation algorithm based on neural network is analyzed first in this article. The shortage of it is easy to fall into local minimum, and it causes the limitation of convergence slowly and separation results inaccurate. Then, a separation algorithm is designed with genetic algorithm and neural network. The algorithm will optimize the initial value of the weight of separation, in order to control the sample by select operation, or cross operation, or mutation operation in the entire search space. Experiments show that it can obtain optimal values for separation matrix, and the speed of audio signal blind separation is quick and effective obviously.