Word-internal context-dependent phoneme models, such as triphones, are used to create Hidden Markov Models (HMMs) for speech recognition. The large number of triphones results the excessive number of model parameters to ...