| 摘要 |
FIELD: information technology. ^ SUBSTANCE: parametres of the input speech signal of a speaker are compared with parametres of a standard speaker selected for comparison by calculating disparity measures - weighted Euclidean distance between parametres of the input and verified speech signals, and the speaker is accepted or rejected based on the calculated disparity measure. For this purpose, values of coefficients of correlation of the input speech signal are determined, values of coefficients of correlation of the mathematical model of the speech signal in form of a multiple carrier frequency pulse of an amplitude-modulated oscillation are calculated, and the minimum distance between coefficients of correlation of the input speech signal and its mathematical mode is calculated, where base frequency is determined and amplitude values of harmonics at the base frequency and overtones of the speech signal, which are the said parametres, are evaluated. Parametres of the input speech signal and the standard can be normalised by dividing the amplitude of carrier harmonics by the standard amplitude-frequency characteristic of the microphone while taking into account the noise level. ^ EFFECT: higher potential reliability of identity verification with high noise immunity when there is noise and when using low-quality microphones. ^ 4 cl, 3 dwg |
