Improvement algorithm of DTW on isolated-word recognition

In order to overcome defects of the search path over tilt in the traditional Dynamic Time Warping (DTW) algorithm and strict alignment at the head and tail of the speech signal, an improved DTW algorithm is proposed to the search path and more relaxed restrictions at the starting point of the speech. Firstly analyzing the traditional DTW, it is found that the traditional search path and strict endpoints restrictions will reduce the success rate of recognition and accuracy. Mel Frequency Cepstrum Coefficients (MFCC) is are used as feature vectors of the reference template and the test template. An optimal match path is chosen through calculating the distortion between the feature vectors with the Euclidean Distance, limiting the slope of the search path from 1/3 to 3 and relaxing speech starting point 3-5 frames. A large number of experiments show that the success rate of recognition is greatly improved in the case of the unchanged identification accuracy.