DEVELOPMENT OF AN ISOLATED WORD RECOGNITION SYSTEM USING MEL FREQUENCY CEPTRUM COEFFICIENT AND DYNAMIC TIME WARPING

Abstract

The recognition accuracy of speech recognition system has been a challenge due to insufficient combination of pre-processing techniques used for speech processing. In order to solve this problem, this system was adopted with well supported pre-processing techniques. Also, in order to enhance the recognition accuracy of the developed speech recognition for this study, it was developed using computational efficient end point detection algorithm that used probability function and linear classifier approach. The development of the speech recognition system was divided into stages. The first stage involved speech recording from different speakers. Three isolated words; count, down and stop, used for the template preparation in this study were taken from ten different speakers using microphone. The second stage involved feature extraction from the recorded speeches using Mel Frequency Ceptrum Coefficients (MFCC). The Third stage was focussed on measuring the global dissimilarity between the stored speech templates and the test input speech samples for the isolated speech recognition using the dynamic time warping algorithm. In the fourth stage, the developed speech algorithm was tested to evaluate its recognition accuracy. The result obtained shows that the developed speech recognition system successfully recognised the isolated words; count, down and stop at different recognition rate of 100%, 60% and 70% respectively. In addition, the result obtained from the developed speech recognition system for this study shows that it performed favourably well in accordance with the global standard for an ideal speech recognition system. This shows that the pre-processing techniques employed in this study indeed enhances the recognition accuracy of the developed speech recognition system. Furthermore, the result obtained from the study shows that the recognition accuracy of the speech recognition system depends on the pre-processing algorithms employed