5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A Comparison of Thai Speech Recognition Systems Using Hidden Markov Model, Neural Network, and Fuzzy-Neural Network

Visarut Ahkuputra, Somchai Jitapunkul, Nutthacha Jittiwarangkul, Ekkarit Maneenoi, Sawit Kasuriya

Department of Electrical Engineering, Chulalongkorn University, Thailand

The recognition of ten Thai isolated numerals from zero to nine and 60 Thai polysyllabic words are compared between different recognition techniques, namely, Neural Network, Modified Backpropagation Neural Network, Fuzzy-Neural Network, and Hidden Markov Model. The 15-state left-to-right discrete hidden markov model in cooperation with the vector quantization technique has been studied and compared with the multilayer perceptron neural network using the error backpropagation, the modified backpropagation, and also with the fuzzy-neural network with the same configuration. The recognition error on Thai isolated numerals using the conventional neural network, the modified neural network, the fuzzy-neural network, and the hidden markov model techniques are 26.97 percent, 22.00 percent, 8.50 percent, and 15.75 percent respectively.

Full Paper

Bibliographic reference.  Ahkuputra, Visarut / Jitapunkul, Somchai / Jittiwarangkul, Nutthacha / Maneenoi, Ekkarit / Kasuriya, Sawit (1998): "A comparison of Thai speech recognition systems using hidden Markov model, neural network, and fuzzy-neural network", In ICSLP-1998, paper 0283.