Spoken Word Recognition Strategy for Tamil Language
S. Palanivel · 2012
This paper outlines a strategy for recognizing a preferred vocabulary of words spoken in Tamil language. The basic philosophy is to extract the features using mel frequency cepstral coefficients (MFCC) from the spoken words that are used as representative features of the speech to create models that aid in recognition. The models chosen for the task are hidden Markov models (HMM) and autoassociative neural networks (AANN). The HMM is used to model the temporal nature of speech and the AANNs to capture the distribution of feature vectors in the feature space. The created models provide a way to investigate an unexplored speech recognition arena for the Tamil language. The performance of the strategy is evaluated for a number of test utterances through HMM and AANN and the results project the reliability of HMM for emerging applications in regional languages.