Application of Mel Frequency Ceptrum Coefficients and Dynamic Time Warping For Developing an Isolated Speech Recognition System

Full Text PDF PDF
Author(s) I.O. Animasahun | J.J. Popoola
Pages 1-8
Volume 4
Issue 1
Date January, 2015
Keywords Speech recognition system, Classes of speech recognition system, End Point Detection Mel Frequency Cepstral Coefficient (MFCC), Dynamic Time Warping (DTW)

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 four stages. The first stage involved speech recording from different speakers. Three isolated words; count, down and stop were taken from ten different speakers using microphone for speech template preparation. 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.

< Back to January Issue