Automatic Estimation of Personality in Children from Spontaneous Voice in Learning Interactions

Resumen

Children's voices convey subtle acoustic cues that reflect paralinguistic information such as individual personality traits, providing a promising, non-intrusive approach for studying socioemotional development in educational contexts. This study investigates the feasibility of estimating personality traits in children aged 8 up to 12 years from spontaneous speech recorded during a programming and problem-solving activity designed as an academic challenge. Personality profiles were obtained using the Children's Personality Questionnaire, a validated psychometric instrument specifically developed for children. A computational pipeline was developed comprising audio segmentation, extraction of acoustic descriptors, feature selection, and stratified cross-validation at the individual level. Several classification paradigms were evaluated, including both multiclass and binary configurations. The best Macro-F1 values were obtained in binary configurations, reaching 0.756 for QI: Adjustment/Anxiety, 0.799 for QII: Introversion/Extraversion, and 0.825 for QIII: Calm/Excitability. However, the QIII result should be interpreted with caution due to the limited representation of minority classes. The obtained findings suggest that children’s voices during challenging learning interactions encode meaningful information that is predictive of personality traits.