ISCA Archive ICSLP 1990
ISCA Archive ICSLP 1990

Neural predictive hidden Markov model

Eiichi Tsuboka, Yoshihiro Takada, Hisashi Wakita

This paper describes a. new type of hidden Markov model where a non-linear predictor composed of a neural network is defined at each state. The idea assumes that the sequence of frames is nonstationary and is a nonlinear autoregressive process whose parameters are controlled by a hidden Markov chain. The parameter estimation methods are shown and the relation of this model with some others is discussed. According to the experimental results, this model shows the best performance among them.