Abstract
Current diagnostic procedures for attention deficit hyperactivity disorder (ADHD) are mainly subjective and prone to bias. While research on potential biomarkers, including EEG, brain imaging, and genetics is promising, it has yet to demonstrate clinical utility. Dopaminergic signaling alternations and executive functioning, crucial to ADHD pathology, are closely related to voice production. Consistently, previous studies point to alterations in voice and speech production in ADHD. However, studies investigating voice in large clinical samples allowing for individual-level prediction of ADHD are lacking. Here, 387 ADHD patients, 204 healthy controls, and 100 psychiatric controls underwent standardized diagnostic assessment. Subjects provided multiple 3-minutes speech samples, yielding 920 samples. Based on prosodic voice features, random forest-based classifications were performed, and cross-validated out-of-sample accuracy was calculated. The classification of ADHD showed the best performance in young female participants (AUC = 0.87) with lower performance in older participants and males. Psychiatric comorbidity did not alter the classification performance. Voice features were associated with ADHD-symptom severity as indicated by random forest regressions. In summary, prosodic features seem to be promising candidates for further research into voice-based digital phenotypes of ADHD.