HIDDEN MARKOV MODEL INCLUDING NEURAL NETWORK FOR SPEECH RECOGNITION

摘要

The hidden markov model (HMM) for processing the time sequence pattern comprises: a model parameter register (2), which stores a parameter of the state transition matrix (A), a parameter of the observation probability density function (B) and a parameter (F) indicating the initial distribution of state; a start model selecting section (1), which satisfys the condition of M0=(A0,B0,0) by selectiong the arbituary parameter (A0,B0,0) from the parameter register; a state sequence separating section (3), which match the syllable signal inputted from the learning data input section (4) with the parameter (A0,B0,0) inputted from the section (1); a parameter reestimation section (5), which looks for a new parameter (An,Bn,Tn).