Joint time-frequency analysis for speech recognition

Abstract

In this paper we present a system developed to recognize a set of words. The feature vectors used are derived from joint time-frequency analysis of the speech signal. We used a multilayer feed-forward neural network trained with the error back propagation algorithm as the feature classifier. We have investigated the network structure that gives the optimum performance. Using a neural network with two hidden layers the system was able to recognize twenty words with 98% recognition rate.