Some tasks connected with speech command recognition on the background of noise

The paper deals with approaches to solve a number of problems which arise in the process of announcer-dependent speech command recognition from a limited dictionary on the back-ground of intensive noise. The recognition is based on representing speech commands in the form of special images - autocorrelation portraits. There appear tasks of metric choice in the space of autocorrelation portraits, taking into consideration pronunciation variability, defining the borders of commands and formation of model library. The algorithms which can help to obtain good ways of solving these problems in acceptable time are suggested.