An efficient method for Tamil speech recognition using MFCC and DTW for mobile applications

Tamil is one of the ancient languages in the world, spoken by 74 million people spread around the world. Tamil is the official language in states like Tamilnadu and countries like Malaysia, Srilanka etc., and the majority of people speak Tamil language. Recognition of Tamil speech would be beneficial to a lot of Tamil people and it is inevitable to carry out research in this field. In this paper we propose a technique for speech recognition which involves preprocessing of signal followed by feature extraction using Mel-Frequency Cepstral Coefficients (MFCC). Mel-frequency Cepstral coefficients (MFCCs) are said to be the coefficients that together represent the short-term power spectrum of a sound, which is based on a linear cosine transform of a log power spectrum on a nonlinear Mel scale of frequency. The process of feature matching is finally carried out using Dynamic Time Warping (DTW). DTW approach is a template matching method, where it stores a prototypical version of each word in the vocabulary called a template and compares the input speech with each word, taking the closest match as matched speech.. In this paper the signal processing techniques, MFCC and DTW are implemented using Matlab and it gives an overview of major technological perspective and appreciation of the fundamental progress of speech recognition.