Automatic Speech Recognition System Using MFCC And DTW For Marathi Isolated Words

Abstract Speech Recognition is the process of identifying words and phrase from spoken language and converts them in to machine readable format. This paper describes an approach of isolated words speech recognition by using Mel-Scale Frequency Cepstral Coefficients MFCC and Dynamic Time Warping DTW. This ASR develops for large vocabulary. MFCC extract the features of spoken words speech signals. For database 100 isolated words are taken from Marathi language and recorded from 100 native speakers with 3 utterances. Dynamic Time Warping algorithm is used for the feature matching purpose. DTW algorithm measures similarity between two sequences which may vary in speed or time.