Parameter influence on speech recognition rate of modified RBF neural network

Abstract

In RBF neural network designing, hidden neuron number and parameter influence the performance of network. The paper discusses influences of pruning hidden neurons using different criteria and parameter on speech recognition rate of modified RBF neural network. First we introduce three hidden neuron pruning criteria, then propose a modified RBF neural network, at last recognition results before and after pruning hidden neuron, relationship among constant C of BIC, hidden neuron number and recognition rates are given. Experiments show that proper criteria and parameter will conduce to better results.