Kovtun V. Informatively-Measuring System of Automatic Speaker Recognition

Dissertation is devoted to the decision of a task of increasing speaker recognition system's efficiency. The purpose succeeded due to the improvement of mathematical model of the speaker's hearing system that allowed formulating the new method of the speaker individual features presentation for his recognition and increasing the right speaker recognition probability on these features. The method of acoustic signal's compression with the adaptive period of sampling is developed. The new method of increasing pitch frequency efficiency for speaker recognition is offered. It allows using as the speaker's individual features the values of wavelet-transformation coefficients on the acoustic signal segments, where the pitch frequency cross-correlation function extreme values are located. On the basis of generalized functionally-statistical criterion the universal method of the efficiency evaluation of speaker recognition system's is developed. Using offered models and methods algorithmic and programmatic facilities are developed. It may be used in speaker recognition systems. Keywords: information-measuring system, speaker, voice, recognition, identification, evaluation, acoustic signal compression.