小波网络和RBF网络的抗噪语音识别

Noise-robust speech recognition based on wavelet network and RBF network

针对目前在噪音环境下语音识别系统性能较差的问题,利用小波神经网络融合了小波变换良好的时频局域化性质和RBF神经网络具有最佳分类能力和辨识能力等特性。构建了一个用小波基替代RBF网络中激活函数的小波-RBF神经网络结构,并采用全监督训练算法,实现了基于小波-RBF网络的抗噪语音识别系统。实验结果表明该系统比RBF网络具有更好的识别效果,尤其在噪声环境下,具有更强的鲁棒性。

To solve the problem that recognition rates of speech recognition systems decrease in the nmsy environment presently, uses character possessing wavelet neural network which integrates the good time-field local property of wavelet transform,uses character possessing RBF neural network,which have best classification ability and recognize ability etc.This paper construetes a wavelet-RBF neural network structure using Morlet mother-wavelet as wavelet basis stead of activate function in the RBF network,adopts whole supervision algorithm and realizes a noise-robust speech recognition system based on wavelet network RBF net-work.The experiment results show that the system has better identify effect than RBF network especially has stronger Robust under noisy environment.