A HMM training algorithm with query-based learning for refinement of classification boundary

A training algorithm of hidden Markov model (HMM) using query-based learning is proposed and applied to the recognition of isolated digits in this paper. An efficient query learning procedure is designed to provide the good training data to the oracle in query-based learning at low cost. The proposed algorithm uses the concept that stems from the gradient based inversion algorithm of artificial neural networks. The proposed algorithm is compared with conventional training methods on isolated digit recognition problem. The results show that the proposed query-based HMM learning algorithm can decrease the recognition error rate up to 60% in our experiments