This paper describes an implementation of the vector quantinizing theory in order 
to use statistic differences of speech elements and make decision towards 
a classification (recognition) of observed elements. Time duration and the 
order of training elements are ignored, while optimality of etalon elements 
depends on  statistic only, and does not depend on temporal rules or other limitation, 
that have effect in the method of recognition which uses a lot of drawings. 
There are given results of a carried out experiments.