Speech recognition using MFCC and DTW classifier

Abstract

Speech recognition had been used broadly in many applications such as security systems, healthcare, and equipment designed for handicapped. This project is about design speech recognition by encoding and modeling the systems in the Digital Signal Processing Toolbox, using two algorithms Mel Frequency Cepstral Coefficients (MFCC) and Dynamic Time Warping (DTW) adapted for feature extraction and classification. First, record the words to accomplish the simulations of the programmed system. An experimental database is obtained by speaking 10 numbers (0-9) during the training phase. Second, that training word has been tested to be matched in order to recognize it. The analysis of coding was modified according to the four elements. They are a number of sample frequency (Fs), types of window used, number of triangles (windowing) and size of the window. From these changes elements we can get the result and determine the best performances of speech recognition. The best performance of this speech recognition using MFCC and DTW algorithms are 90% recognition rate. Thus, the designed systems actually work well in the speech recognition.