The Performance Comparison between Backpropagation and Extended Incremental Network Model on Speaker Verification

摘要

In this paper we use a neural network algorithm for speaker verification. A scheme for finding a solution to problems of incremental learning algorithm is used when the structure becomes complex and noise patterns are shown in the learning data set. Our approach for this problem uses a pruning method that will terminate the learning process with a predefined criterion. We compare the performance of Backpropagation and extended incremental network model learning algorithms on speaker verification. The network is retrained using Backpropagation to show the effectiveness of the pruning method. The incremental network model with Fisher's linear discriminant function demonstrates its superiority for speaker verification than backpropagation.