Abstract：The approach for speaker recognition based on neural networks is able to emulate the function of human brain in some degree, so it is a main implementation technology in the speaker recognition. But it is difficult to determine the number of hidden layer neurons, slowly convergent and easy to fall into local minimum point. The model of wavelet neural networks is studied. The structure of the network and learning algorithm are given. The recognition correctness reaches to 99.5% for 5 speakers using Mel frequency cepstral coefficient as feature parameters. The experimental at results show that the learning rate and recognition correctness are improved much compared to the BP networks. It has a good application prospect and worth to research further more.
2006, Vol. 28 
