Автентифікація клієнтів за голосовим відбитком в режимі реального часу

Abstract

An analysis of modern client authentication approaches in real time was carried out and a text-independent voice authentication approach was chosen. Modern information technologies of client authentication in real time (BioLink AMIS, Voice Key and VocalPassword) were analyzed and their main drawback was determined - the insufficient speed of the authentication process, which does not allow client authentication in real time and limits the scope of use. An analysis of voiceprint classifiers for real-time customer authentication was conducted and K�NN was ed as the most accurate among those analyzed. The choice of the mathematical method MFCCs (mel-cepstral coefficients) for voice authentication of the client is justified, which will increase the speed of the authentication process in combination with the ed K-NN (K - nearest neighbors) classifier. The mathematical model of the real-time client authentication process has been improved due to the introduction of the authentication factor in the cepstral analysis of voice prints. The real-time customer authentication process method is improved, which is based on the combination of the MFCCs mathematical method for voice prints and the K-NN classifier of voice prints. An improved real-time client authentication information technology is proposed, which provides the analysis of the client`s voice prints and implements an improved method of real-time client authentication, which made it possible to increase the speed of the authentication process and not deteriorate the accuracy of the client`s authentication