Keyword spotting system for Tamil isolated words using Multidimensional MFCC and DTW algorithm

Audio mining is a speaker independent speech processing technique and is related to data mining. Keyword spotting plays an important role in audio mining. Keyword spotting is retrieval of all instances of a given keyword in spoken utterances. It is well suited to data mining tasks that process large amount of speech such as telephone routing and to audio document indexing. Feature extraction is the first step for all speech processing tasks. This Paper presents an approach for keyword spotting in isolated Tamil utterances using Multidimensional Mel Frequency Cepstral Coefficient feature vectors and DTW algorithm. The accuracy of keyword spotting is measured with 12D, 26D and 39D MFCC feature vectors for month names in Tamil language and the performances of the mu1tidimensional MFCCs are compared. The code is developed in the MATLAB environment and performs the identification satisfactorily.