Mathematical Methods Application in an Artificial Neural Network to Recognize "Fake" Emotions in the Voice

This paper proposes a methodology for comparing the emotional component of the voices of speakers actually experiencing emotions and those trying to reproduce them. The technique assumes the use of as classifiers the ratio of frequencies of local maxima of the spectrum of the of a sound wave.  In this work we studied 4 categories of basic emotions: anger, sadness, fear, joy. To obtain comparative characteristics in each category 30 records with "fake" and " true" emotion were investigated – 240 records in total. A statistical comparative analysis of the classifiers was performed and found a significant differences in the data for the genuine emotion and its imitation.  Also, a model of an artificial neural network, based on which a program was created to recognize the emotional message contained in human voice recordings. On the trained artificial neural network, an experiment was conducted to determine the emotional state of the speaker and the truth or falsity of his emotional message.  Testing of sound files with different emotions showed good results for recognizing both the emotional state of the speaker and the truthfulness or falsity of the speaker's emotional message.