Abstract. The paper presents the methodological foundations necessary to solve the problem of automatic classification of emotions by the speaker’s speech. The general principle of synthesis of the emotion classification algorithm based on machine learning methods is proposed. A model of the process of speech formation is presented, which allows using digital signal processing tools to isolate significant informative features from speech. The general process of constructing a mathematical model of a classifier based on a probabilistic approach is described.

Keywords: emotion recognition, speech, automatic classification, machine learning.