Чубатый Д.Н. «Эталонная модель источника речевых сообщений на основе гауссовских смесей» Наукоемкие технологии, 13, № 8, с. 23-27 (2012)

Предложена эталонная модель источника речевых сообщений на основе Гауссовых смесей для идентификации личности в сетях сотовой связи стандарта GSM. Показано, что особенностью этой модели является использование вектора различительных признаков, сформированного из параметров, передаваемых в кадре низкоскоростного кодера речи RPE-LTP. Determine the personality of the speaker system on the individual characteristics of the speech developing very actively. This is due to the presence of a wide range of practical problems which can be used in these systems: verification of access rights, forensic examination, remote access to databases and the bank accounts. Recently, large distribution network were using low speed speech coding: satellite, trunking, cellular communications systems, as well as IP-telephony network. The application of existing method of identification person by voice in these networks is not possible. This requires the development of new models, method sand algorithms for speech signal processing, low-speed converted speech coder. Presented in the paper reference model of the source voice messages based on Gaussian mixtures confirmed the possibility of using parameters that are passed within the frame coder RPE-LTP as a vector of distinctive features and identity of the caller without decoding the signal.