Affective Dimensions in Maternal Voice During Child Feeding in Mothers With and Without Eating Disorder History-Findings From a Machine Learning Analysis of Speech Data.

Abstract

Objective: Eating disorder (ED) history may impact mother-child communication during mealtimes and contribute to transgenerational transmission of ED. This study employed machine learning (ML) to identify speech characteristics during mother-child feeding interactions, aiming for investigating whether vocalised affective characteristics differ between mothers with and without ED history when feeding their child.

Method: Mothers with (n = 17) and without ED history (n = 27) and their children (10 months) were filmed at home during mealtime. Various ML models were exploratively tested to assess their suitability for analysing maternal voice data. Diagnosis of an ED history was based on the structured Eating Disorder Examination Interview.

Results: A ML model specialised for the prediction of emotional arousal, valence and dominance provided the most pronounced differences between the groups. These variables were consistently stronger expressed in the voices of mothers with ED history during child feeding, predominantly in the middle of the interaction.

Conclusions: Voice data suggests that mothers with ED history might be emotionally stronger involved throughout child feeding. This indicates that there are differences in communication between women with and without ED history and highlights the importance of research into maternal communication in affected families. ML approaches are promising tools as they can detect more subtle nuances compared to questionnaires.