Comparative Analysis of Authentication Using Formant Features of Vowels and Consonants

This article discusses 3 different biometric images of the user, which are based on the formant characteristics of the sounds of the Russian language. In the 1st biometric image, the formant characteristics of vowels (including diphthongs) and consonants are used, in the 2nd—vowels and diphthongs, and in the 3rd—consonants. A cluster of algorithms is presented, consisting of the Random Forest algorithm, K Nearest Neighbors, Logistic Regression, Multilayer Perceptron and Support Vector Machines. On the basis of a cluster of algorithms, models have been developed and trained. The mechanism of creation and training of each model is considered. To compare the effectiveness of user biometric images, an experiment was conducted in which 10 people participated. In total, users pronounced 46 sounds (6 vowels, 4 diphthongs and 36 consonants). Based on the provided data, samples were formed for each biometric image. Each sample, when it enters the input of a cluster of algorithms, is divided into training and test. A comparative analysis of the authentication procedure with each biometric image was carried out. For this, the following parameters were used: the average calculation time, the average accuracy value and the average accuracy value on the test sample.