This paper presents a hybrid framework of feature extraction and hidden Markov modeling (HMM) for two-dimensional pattern recognition. Importantly, we explore a new discriminative training criterion to assure model compactness and discriminability. T

this Program utilized a new method for generation codebook in Hidden Markov Models (HMM). Hybrid HMM and neural networks have been applied to improve recognition in speech processing. -this Program utilized a new method for generation