Reliably quantifying the severity of social symptoms in children with autism using ASDSpeech

Abstract

Several studies have demonstrated that the severity of social communication problems, a core symptom of Autism Spectrum Disorder (ASD), is correlated with specific speech characteristics of ASD individuals. This suggests that it may be possible to develop speech analysis algorithms that can quantify ASD symptom severity from speech recordings in a direct and objective manner. Here we demonstrate the utility of a new open-source AI algorithm, ASDSpeech, which can analyze speech recordings of ASD children and reliably quantify their social communication difficulties across multiple developmental timepoints. The algorithm was trained and tested on the largest ASD speech dataset available to date, which contained 99,193 vocalizations from 197 ASD children recorded in 258 Autism Diagnostic Observation Schedule, 2^nd edition (ADOS-2) assessments. ASDSpeech was trained with acoustic and conversational features extracted from the speech recordings of 136 children, who participated in a single ADOS-2 assessment, and tested with independent recordings of 61 additional children who completed two ADOS-2 assessments, separated by 1–2 years. Estimated total ADOS-2 scores in the test set were significantly correlated with actual scores when examining either the first (r(59) = 0.544, P < 0.0001) or second (r(59) = 0.605, P < 0.0001) assessment. Separate estimation of social communication and restricted and repetitive behavior symptoms revealed that ASDSpeech was particularly accurate at estimating social communication symptoms (i.e., ADOS-2 social affect scores). These results demonstrate the potential utility of ASDSpeech for enhancing basic and clinical ASD research as well as clinical management. We openly share both algorithm and speech feature dataset for use and further development by the community.