Automatic children’s personality assessment from emotional speech

Abstract Identifying personality traits in children is a topic of interest due to its importance in adapting strategies for the teaching–learning process and detecting psychopathological features. The most straightforward procedure to identify children’s personality type is by applying a validated and suitable questionnaire according to their age. However, an interesting approach is automatically identifying the children’s personalities using their speech generated during their interaction with computer systems, software, or robots. This approach would allow obtaining the personality identification transparently to the children without answering a written test. This article presents a method for the automatic personality assessment of children between 8 and 12 years old. The assessment is based on their voices’ acoustic analysis while participating in playful activity with another child and a robot. We created a database with 98 children involved in several activities while their voices were recorded. The database was labeled with five paralinguistic aspects. Using these labels, we trained a set of classification models that helped us recognize primary and secondary personality traits according to the Children’s Personality Questionnaire. We obtained good results for two secondary personality traits, extroversion (0.89 F-Score) and excitability (0.79 F-Score). The best F-score obtained for anxiety was 0.70. These results indicate that it is feasible to estimate personality from analyzing children’s voices during interaction with computer systems.