International Journal of Computer Theory and Engineering

Abstract

Speech Recognition by computer is a process where speech signals are automatically converted into the corresponding sequence of words in text. When the training and testing conditions are not similar, statistical speech recognition algorithms suffer from severe degradation in recognition accuracy. So we depend on intelligent and recognizable sounds for common communications. In this research, word inputs are recognized by the system and executed in the form of text corresponding to the input word. In this paper, we propose a hybrid model by using a fully connected hidden layer between the input state nodes and the output. We have proposed a new objective function for the neural network using a combined framework of statistical and neural network based classifiers. We have used the hybrid model of Radial Basis Function and the Pattern Matching method. The system was trained by Indian English word consisting of 50 words uttered by 20 male speakers and 20 female speakers. The test samples comprised 30 words spoken by a different set of 20 male speakers and 20 female speakers. The recognition accuracy is found to be 91% which is well above the previous results.