摘要 |
PURPOSE: To reduce the calculation quantity of likelihood and to adapt a small amount of learning data. CONSTITUTION: Hidden Markov models A and B have single distributions, for example, in a state 1, and the mean vectors of the distributions of the models A and B are denoted as four-dimensionalμa =(μa ,1 ,μa ,2 ,μa ,3 ,μa ,4 ) andμb =(μb ,1 ,μb ,2 ,μb ,3 ,μb ,4 ). Then nearly equal corresponding elements of those mean value vectors, e.g.μa ,2 andμb ,2 are regarded as a common elementμc ,2 . For the likelihood calculation of the models A and B for an input vector Xt =(xt ,1 , xt ,2 , xt ,3 , xt ,4 ), the arithmetic result of (xt ,2 -μc ,2 )<2> is used in common instead of (xt ,2 -μa ,2 )<2> and (xt ,2 -μb ,2 )<2> .
|