WORD CATEGORY PREDICTION BASED ON NEURAL-NETWORK

This paper proposes a new method of the word category prediction for t he speech recognition system. In order to improve the speech recogniti on results, not only the acoustical information but also certain lingu istic information is needed. World category prediction is a very effec tive method to implement an accurate word recognition system. Traditio nal statistical approaches require considerable training data to estim ate the probabilities of word sequences, and many parameters to memori ze probabilities. And it is difficult to predict unseen data which doe s not include the training data. To solve this problem, NETgram, which is the neural network for word category prediction, is proposed. The performance of the NETgram is comparable to that of the statistical mo del although the NETgram requires fewer parameters than the statistica l model. Also the NETgram performs effectively for unknown data, i.e., the NETgram interpolates sparse training data. Results of analyzing t he NETgram show that the NETgram learns linguistic structure From trai ning data. The results of applying the NETgram to HMM English word rec ognition show that the NETgram improves the word recognition rate from 81.0% to 86.9%.