改进粒子群算法的小波神经网络语音去噪

Speech Denoising Based on the Wavelet Neural Network of Improved Particle Swarm Algorithm

小波神经网络是小波分析和神经网络相结合的产物,它结合了小波分析良好的时频局域化性质和神经网络自学习功能,因而使得小波神经网络具有较强的逼近和容错能力,并具有良好的收敛性和鲁棒性;充分利用小波神经网络的优点,提出一种改进的粒子群优化算法,通过个体间的协作与竞争寻找最优解,并将其应用于语音去噪的研究;最后通过Matlab仿真结果表明,将粒子群优化算法融合到小波神经网络对语音信号进行消噪是一种有效的方法,它能在一定程度上去除噪声,使原信号的特征尖峰点得到了很好的保留,更好地估计原始信号,明显地改善了语音增强的效果。

Wavelet neural network is the combination of the wavelet analysis and neural network, it combines good timefrequency lo calization properties of wavelet analysis and neural networks selflearning function. Thus wavelet neural network has strong approximation and fault tolerance, and has good convergence and robustness. This paper makes full use of the advantages of wavelet neural network, puts forward an improved particle swarm optimization algorithm, which applies to speech denoising research through collaboration and competition among the individuals to find the optimal solution. Finally, the Matlab simulation results prove that the particle swarm optimization algo rithm is integrated into wavelet neural enhancement network is a kind of effective method to speech signal denoising , it can remove the noise to some extent, makes the characteristic peak points of original signal are well preserved and estimate the original signal preferably, this method obviously improves the speech enhancement effect.