A speech recognition system for Malaysian English pronunciation using Neural Network

Abstract

The English language as spoken by Malaysians varies from place to place and differs from one ethnic community and its sub-group to another. In this paper, an automatic vowel classification system based on linear predictive coding (LPC) and neural network is presented to understand the English pronunciation as spoken by Malaysians. A database consisting of 11 words recorded from 10 speakers is created and used in this work. The input signal is pre-emphasised and frames features are extracted using LPC; a simple feedforward neural network trained by conventional backpropagation procedure in four different modes of activation functions is also proposed. To stabilize the cumulative error versus epoch training and to minimize the training time, a systole activation function is also proposed. The results obtained from the neural network trained by systole activation function are compared with the sigmoidal activation functions.