An Efficient Implementation of Speech Recognition based on Curvelet Transform and Artificial Neural Network

Abstract

Speech Recognition is ability to translate a dictation or spoken words to text format. In the field of electronics and computers, speech has not been used much more due to the complexity and different types of sounds and speech signals. However, with traditional methods, processes and algorithms, we can simply process the speech signals and identify the text. This paper presents an efficient speech recognition system based on discrete curvelet transform (DCT) and Artificial Neural Network (ANN) methods to enhance the identification rate. This research article comprised in two distinct phases, a feature extractor and a recognizer is presented. In Feature Extraction phase, Curvelet transform extract the features called curvelets from the given input speech signal and elements of these signals which support in gaining higher recognition rates. For feature matching, Artificial Neural Networks is used as classifiers. The performance evaluation has been demonstrated in terms of accurate recognition rate, maximum noise power of interfering sounds, miss rates, hit rates, and false alarm rate. The accurate classification rate was 98.3 % for the sample speech signals. Performance comparisons with similar studies found in the related literature indicated that our proposed ANN structures yield satisfactory results and improve the recognition rates.