F021 Acoustic characterization of Huntington’s disease emotional expression: an explainable AI approach

<h3></h3> Huntington’s Disease (HD) is a neurodegenerative disorder characterized by motor, cognitive, and psychiatric symptoms. The first studies on the emotional behavior of HD patients, based on speech emotion recognition (SER) models, have recently emerged. Following on from this line of research, we propose to study two unexplored aspects: 1. The ability of HD individuals to express emotions using emotional vocal narration; 2. The nuances of acoustic expression across various emotions in HD patients. 73 individuals with manifest HD, 28 premanifest HD (preHD) and 37 healthy controls (HC) were recorded when prompted to narrate happy, angry, sad and neutral stories. These emotional stories recordings were independently assessed by 5 raters for their emotional content on Likert-scales, to analyze the groups’ ability to follow the given instructions. Based on these recordings’ emotional ratings, speech emotion classifiers were developed using Random Forest, to which explainable AI methods (Shapley Additive exPlanations (SHAP)) were applied to dissect each group’s acoustic emotional characteristics. Compared to controls, HD individuals exhibited difficulties in narrating happy and angry stories, whereas preHD only showed difficulties in narrating angry stories. The interpretation of the SHAP analysis showed that the important acoustic features in the prediction of emotions differed from one group to another (figure 1). Overall, this contributes to a deeper understanding of emotional speech narration in HD and preHD individuals, highlighting not only the deviations from controls but also uncovering specific acoustic patterns indicative of emotional expression in the HD group.