DEVELOPMENT OF AN ALGORITHM FOR CLASSIFYING THE EMOTIONAL STATE OF THE SUBJECT BASED ON SPEECH DATA USING THE MATLAB PACKAGE

Semenyuk Victoria Valeryevna, Skladchikov Maxim Vladimirovich

The purpose of the work is to analyze current models, approaches and algorithms of image recognition systems. The implementation of a software package based on convolutional neural networks in order to improve the quality of pattern recognition and the construction of an optimal algorithm for recognizing objects based on the received speech data. Computer pattern recognition is a rather complex task, which is successfully solved through the use of artificial neural networks. Automatic identification of images (text, sound, face, objects, etc.) using a computer is one of the most promising areas of development of artificial intelligence technologies, which allows us to give a key to understanding the features of human intelligence. To classify the emotional state of the subject, an algorithm was created using convolutional neural networks. For training, a specialized data set was selected, presented in the form of audio recordings characterizing a separate emotional state of the subject. The input of the trained model received transformed speech data in the form of a spectrogram image containing characteristic signs of a certain emotion. As a result of the study, an algorithm for classifying the emotional color of the subject based on speech data was developed. Experimental studies were carried out to assess the accuracy of the algorithm. This study is a logical continuation of the work in which a similar algorithm was also created [1]. However, due to the low accuracy of the work, it was necessary to improve it. As a result, it was decided to create a different neural network structure by means of the Matlab package. The proposed algorithm, due to its own universality, can be applied in various fields for the tasks of recognizing the emotional color of the subject.

Keywords:neural network, human emotion recognition, convolutional neural network, sound fingerprinting, Tenserflow, Matlab