Abstract
Current practices in the assessment and diagnosis of autism spectrum disorder (ASD) rely on expert clinician observation and judgement, but objective measurement holds the potential to supplement and support clinicians in this process. The current study utilized machine learning and audio signal processing to identify objective measurements of vocal behaviors and proxemics which could predict clinician-rated indices of autism symptom severity. Video and audio recordings of administration of the Autism Diagnostic Observation Schedule (ADOS-2) were taken and analyzed for 66 preschool age children who were evaluated for ASD at a university-based clinic. Several indices of child and adult movement within the assessment room were associated with clinician-rated Social Affect symptom severity. Vocalizations of children and adults predicted clinician-rated Restricted and Repetitive Behavior symptom severity. Implications for the use of objective measurement in ASD assessment and diagnosis are discussed.