A problem of speech recognition is presented as a problem of combinatorial optimization. 
It is shown that for it an objective function depends on several variables, 
which are the combinatorial configurations of different types: the n-element 
set partitions into subsets, the placements, the combinations without repetitions. 
For these definitions, the problem of speech recognition is divided into three subproblems. 
For a decision of each subproblem, an independent algorithm is constructed.
A scheme of the structurization of a library of standard signals is described. 
The use of this structurization permits the dictionary look-up of the word 
with direction searching. The problem of speech recognition is conducted 
by hybrid algorithm, which is based on organization of iterative process 
which generates the sequence of solutions in accordance with built-in algorithms.