MULTIPLE EFFECTIVENESS CRITERIA OF FORMING DATABASES OF EMOTIONAL VOICE SIGNALS

Abstract

A significant number of created databases of emotional speech in different languages testifies to the great interest of the research community in the problems of synthesis of emotional voice signals and recognition of emotions in the human voice. In our time, devices that use a voice interface for user interaction are gaining significant use, which is especially important in certain robotic systems.

As a basis for the creation of computer systems for recognizing emotions in a person's voice, neural networks are usually used, and for their training, sufficiently large databases of emotional voice signals are required. The main approach used in the creation of such databases is the involvement of actors to reproduce a given range of emotions in their utterances, and, accordingly, the use of specialized equipment for recording and analyzing the received audio data. However, this approach requires significant time and resource costs, which does not allow generating significant volumes of emotional voice expressions in a reasonable period of time.

Therefore, in order to evaluate the effectiveness of the formation of databases of emotional voice signals, a list of criteria is given, according to which the means of forming emotional databases were evaluated. The results of the evaluation allow us to reasonably claim that the known means of forming emotional databases of human voice signals have a certain number of shortcomings. In order to increase the efficiency of the means of forming databases of emotional voice signals, it is advisable to provide the possibility of forming databases without the involvement of professional actors, the presence of spontaneous expressions, not only predetermined ones, the presence of polyphonic expressions, namely dialogues, and the presence of opportunities for evaluating time and computing resources, which are necessary for the formation of database elements.