Спектральные характеристики сигналов для их классификации по уровню раздражительности методами машинного обучения

Abstract

The popularity of using machine learning methods to classify audio signals continues to grow. Although the recognition of qualitative parameters of noise is becoming increasingly important, its negative aspects have been rarely studied. Methods for determining the level of noise annoyance are presented. These methods can be useful for creating systems that control the noise level of the environment, as well as for evaluating not only acoustic characteristics, but also aesthetic and cognitive effects exerted by noise on a person. A comparison of the results of a neural network in determining the degree of noise annoyance in terms of the psychoacoustic parameter using various spectral characteristics is presented.