Two-Layer Perceptron for Voice Recognition of Speaker’s Identity

The article is devoted to the problem of ensuring reliable authentication of users of information systems for various purposes. The prospects of solving this problem through the use of voice signal analysis tools to recognize the speaker’s personality are shown. The main advantages of such tools include the increased durability of the biometric access code, the use of common registration tools, as well as the possibility of implementation of hidden monitoring of the user’s identity. The relevance of research in the direction of developing low-resource means of recognizing the speaker’s personality by voice fragments of a fixed duration, using only available computing power on the spot, is substantiated. Based on the analysis of literary works, the prospects of using neural network solutions are shown, the creation of which is complicated by the existing uncertainty in choosing the type of neural network model, as well as in determining the set of input parameters. As a result of the studies, it was determined that in the task of recognizing the speaker’s identity by voice fragments of a fixed duration, it is advisable to use a type of neural network model such as a two-layer perceptron, the input parameters of which are associated with small-cepstral coefficients characterizing each of the quasi-stationary fragments of the analyzed voice signal, and the output parameters match of recognizable speakers. By computing experiments, it is proved that each of the quasistationary fragments should be described using 20 chalk-cepstral coefficients. At the same time, the recognition accuracy of the speaker using a two-layer perceptron is at the level of the best modern means of this purpose and is 8% higher than the recognition accuracy using a convolutional neural network such as LeNet. The need for further research in the direction of adapting the parameters of the two-layer perceptron to the recognition conditions under the influence of various kinds of interference was also established.