期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2019
卷号：97
期号：22
页码：3344-3355
出版社：Journal of Theoretical and Applied
摘要：This article introduces a comparison between two different techniques for the selection of speech features. These features can be used for speaker recognition or speech recognition. Feature selection is very effective for recognition accuracy. A comparison between short time energy per frame and the mel frequency cepstral coefficient (MFCC) to extract the speech features is given. Neural network and Hidden Markov model are used as classifier tools. The objective of the article is to enhance the recognition rate of phonetic Arabic letters through selecting the proper speech feature. Dynamic Time Warping (DTW) technique is used to align the analogous frames from different samples of the same signal. An effective and robust method is proposed to evaluate the feature of spoken Arabic letters. This work introduces applying the Mel Frequency Cepdstral coefficient(MFCC) to extract the speech feature.The objective of the proposed system is to enhance the performance by introducing three systems which are proposed to recognize the spoken Arabic letters. The first is based on neural networks. The second is based on hidden Markov model. Third system is based on combination between neural networks and hidden Markov models. The accuracy of neural network is found to be 42% with MFCC for 84 spoken letters while with short time energy it is found to be 84.3% for 28 spoken letters. By grouping the letters into similar letters, the accuracy of feature based on short time energy reached to 98.9%.For MFCC, the hidden Markov model performance is found to be 98.5%. But for combination system based on neural network and hidden Markov models with MFCC, the accuracy of 99.25% is obtained.
