A hybrid RBF-HMM system for continuous speech recognition

A hybrid system for continuous speech recognition con sisting of a neural network with Radial Basis Functions and Hidden Markov Models is described in this paper together with discriminant training techniques Initially the neural net is trained to approximate a posteriori probabilities of single HMM states These probabilities are used by the Viterbi algorithm to calculate the total scores for the indi vidual hybrid phoneme models The nal training of the hy brid system is based on the Minimum Classi cation Error objective function which approximates the misclassi ca tion rate of the hybrid classi er and the Generalized Pro babilistic Descent algorithm The hybrid system was used in continuous speech recognition experiments with phoneme units and shows about phoneme recognition rate in a speaker independent task