Quantifying vocal subsystems of adults with minimally-verbal Autism Spectrum Disorder

Thomas Quatieri^1*Tanya Talkar¹James Williamson¹Sophia Yuditskaya¹Chakameh Jafari¹Meredith Pecukonis²Paige Hickey Townsend²Laura Sarnie²Nataliya Kosmyna³Nishat Protyasha³Pattie Maes³Christopher McDougle²Lisa Nowinski²Maria Mody²

• ¹Lincoln Laboratory, Massachusetts Institute of Technology, Lexington, United States
• ²Massachusetts General Hospital Lurie Center for Autism, Lexington, United States
• ³Massachusetts Institute of Technology Media Lab, Cambridge, United States

The understanding of speech production and its fine-motor underpinnings in individuals with autism spectrum disorder (ASD) has become an area of growing interest in clinical research. Here, we developed objective acoustic-based measures to characterize and contrast the respiratory, laryngeal, and articulatory vocal production subsystems based on data from 27 adults with minimally-verbal autism spectrum disorder (mv-ASD) relative to 27 age-matched neurotypical (NT) peers. The complexity of representative movement dynamics of each subsystem was determined through signal correlation analysis, associated with degrees of freedom derived from correlation structure. As proxies to underlying speech production, we used base features of signal envelope, pitch and formant trajectories associated with each subsystem, respectively, along with their velocities. In addition, we examined cepstral peak prominence and mel-frequency cepstral coefficients, associated with laryngeal and articulatory systems, widely used in assessing voice and speech in neurological and pathological conditions. With speech data from a protocol consisting of diadochokinetic sequencing (DDK), and three different single-word tasks of varying cognitive load (Imitation, Naming, Reading), we studied complexity of motor dynamics at group and individual levels under each task condition. At group level, mv-ASDs showed lower complexity than NT participants for the respiratory and laryngeal subsystems and, except for the DDK task, higher complexity for the articulatory subsystem. At the individual level, there is a range of complexity consistent with mv-ASD heterogeneity. Using Pearson and Spearman correlation measures, we correlated our subsystem characterization with measures of non-verbal IQ, expressive and receptive vocabulary, visual-motor integration, and fine motor dexterity to understand the clinical relevance of these acoustic features, as well as the effect of outliers. When combining mv-ASD and NT groups, articulatory base features were most consistent in correlating with expressive vocabulary across tasks, compared to respiratory and laryngeal features. The Pearson-based association holds within the mv-ASD group alone for the single-word tasks. Continued understanding of speech and language challenges of individuals with mv-ASD reflected in metrics of each of the three speech production subsystems (respiratory, laryngeal, and articulatory) and their relationship with standardized motor, cognitive, and language assessments, will promote design of functionally-tailored personalized and clinically acceptable interventions.