Machine learning model for reproducing subjective sensations and alleviating sound-induced stress in individuals with developmental disorders.

Abstract

Introduction: An everyday challenge frequently encountered by individuals with developmental disorders is auditory hypersensitivity, which causes distress in response to certain sounds and the overall sound environment. This study developed deep neural network (DNN) models to address this issue. One model predicts changes in subjective sound perception to quantify auditory hypersensitivity characteristics, while the other determines the modifications needed to sound stimuli to alleviate stress. These models are expected to serve as a foundation for personalized support systems for individuals with developmental disorders experiencing auditory hypersensitivity.

Methods: Experiments were conducted with participants diagnosed with autism spectrum disorder or attention deficit hyperactivity disorder who exhibited auditory hypersensitivity (the developmental disorders group, DD) and a control group without developmental disorders (the typically developing group, TD). Participants were asked to indicate either "how they perceived the sound in similar past situations" (Recollection task) or "how the sound should be modified to reduce stress" (Easing task) by applying various auditory filters to the input auditory stimulus. For both tasks, the DNN models were trained to predict the filter settings and subjective stress ratings based on the input stimulus, and the performance and accuracy of these predictions were evaluated.

Results: Three main findings were obtained. (a) Significant reductions in stress ratings were observed in the Easing task compared to the Recollection task. (b) The prediction models successfully estimated stress ratings, achieving a correlation coefficient of approximately 0.4 to 0.7 with the actual values. (c) Differences were observed in the performance of parameter predictions depending on whether data from the entire participant pool were used or whether data were analyzed separately for the DD and TD groups.

Discussion: These findings suggest that the prediction model for the Easing task can potentially be developed into a system that automatically reduces sound-induced stress through auditory filtering. Similarly, the model for the Recollection task could be used as a tool for assessing auditory stress. To establish a robust support system, further data collection, particularly from individuals with DD, is necessary.