Ensemble of Logistic Regression Models for Voice Settings of Virtual Pedagogical Agent

The use of virtual pedagogical agents in a digital learning environment can improve learning efficiency. To do this, when developing, it is necessary to consider and lay down the parameters that will affect the interaction between the agent and the student. The agent’s voice is an important form of incarnation. Voice quality affects learning, perception, and trust. That is why it is so important to set voice parameters. Since each student has his characteristics, models are needed that consider such properties when setting voice parameters. However, such models are rare in the literature. The work aims to develop models for the voice settings of a virtual pedagogical agent that considers the individual properties of a particular student. Since an important parameter of speech is the tempo, which affects the comfort and convenience of listening, the tempo of the agent’s speech is chosen as a dependent variable. As independent variables, indicators of the strength of the student’s nervous system were used, which determine the course of neurodynamic and cognitive processes that affect the effectiveness of learning. An experiment was carried out, and based on the results of processing the data obtained, an ensemble of logistic regression models was built that determines the rate of speech according to the selected indicators of the strength of the nervous system.