AUTOMATIC-DETERMINATION OF THE NUMBER OF MIXTURE COMPONENTS FOR CONTINUOUS HMMS BASED ON A UNIFORM VARIANCE CRITERION

We discuss how to determine automatically the number of mixture compon ents in continuous mixture density HMMs (CHMMs). A notable trend has b een the use of CHMMs in recent years. One of the major problems with a CHMM is how to determine its structure, that is, how many mixture com ponents and states it has and its optimal topology. The number of mixt ure components has been determined heuristically so far. To solve this problem, we first investigate the influence of the number of mixture components on model parameters and the output log likelihood value. As a result, in contrast to the ''mixture number uniformity'' which is a pplied in conventional approaches to determine the number of mixture c omponents, we propose the principle of ''distribution size uniformity' '. An algorithm is introduced for automatically determining the number of mixture components. The performance of this algorithm is shown thr ough recognition experiments involving all Japanese phonemes. Two type s of experiments are carried out. One assumes that the number of mixtu re components for each state is the same within a phonetic model but m ay vary between states belonging to different phonemes. The other assu mes that each state has a variable number of mixture components. These two experiments give better results than the conventional method.