Speech Recognition Approach using Descend-Delta-Mean and MFCC Algorithm

In this paper, we present the improving the feature extraction method of the MFCC algorithm for noise robustness when there is interference in the speech input signal. The various parameter extraction with new technique as the descend-delta-mean and mel frequency cepstral coefficients is called the DDMMFCC. The purpose is to require the noise robustness of feature vector extracted from the speech signal with the characteristic coefficients of the DDM-MFCC algorithm. In order to increase the speech and speakers recognition accuracy rate, the DDM technique is a new method of finding the average value of a descendant of power spectrum magnitude. The process consisting method of rearranging magnitude of the spectral power is reduced to a minimum in each frequency band of the speech signal, and to find the difference in the arrangement. Finally, to determine the value of mean. The experiment results of the proposed robustness feature extraction DDM-MFCC algorithm and the conventional of MFCC algorithm is compared, when the training and testing dataset is the 0-9 Isolated THAI digits speech of 10 THAI speakers recognition. With additive random white Gaussian noise is added for testing dataset with signal to noise ratio (SNR) as 0, 5, 10, 15 and 20 dB, respectively. Simulation results of proposed DMM-MFCC algorithm are shown higher rate of recognition accuracy rate than the conventional MFCC algorithm.