Neural Network Models for Recognition of Consonant-Vowel ( C V) Utterances

I n this paper, we present an approach based o n neural network models for recognition of utterances of syllable-like units in Indian languages. The distribution captur-ing ability of a n autoassociative neural network model is exploited to perform nonlinear principal component analysis f o r compressing the size of the feature vector. A constraint satisfaction model is proposed t o incor-porate the acoustic-phonetic knowledge and to combine the outputs of subnets t o arrive at the overall decision on the class of a n input utterance.