Exploring motor speech patterns in adults with minimally verbal autism spectrum disorder through surface electromyography.

Introduction: It is well known that standardized neuropsychological testing frequently fails to capture the true capacity and full range of abilities in individuals with minimally verbal autism spectrum disorder (mvASD) due to difficulties with motor speech skills. Here we used Surface Electromyography (sEMG), a non-invasive method that captures action potentials during muscle movements, to examine the motor basis of speech production challenges in adults with minimally verbal autism spectrum disorder (mvASD). Method: sEMG data were collected from 8 sensors placed on the face and neck while participants performed four speech tasks: imitation, naming and reading words, and a syllable repetition task (diadochokinetic, DDK). We compared adults with mvASD and neurotypical controls (NT) on RMS amplitude and mean correlation between the sEMG signals and movement complexity measures derived from auto and cross-correlation structures of the signal to capture the dynamic coordination of muscle activity during speech production. Principal component analysis was applied to reduce the eigenvalue features to a compact speech-motor representation used as input to leave-one-out cross-validated predictive models of group. Results: Across all speech tasks, the mvASD group consistently demonstrated stronger correlations between sensor signals from muscles on the face and neck compared to NT. The finding suggests tighter coupling of muscle activity reflecting potentially less differentiated muscle movement control during speech production. This is in keeping with a pattern of lower complexity of motor coordination related to reduced degrees of freedom in mvASD compared to NT participants. Conclusion: The findings extend previous work by demonstrating that sEMG features capture differences in muscle coordination patterns between adults with mvASD and neurotypical individuals across multiple speech tasks. Additionally, dimensionality-reduced EMG representations derived from imitation and diadochokinetic (DDK) tasks showed the strongest ability to distinguish between groups, suggesting that speech tasks that involve reproducing the utterances presented may be particularly sensitive to atypical speech-motor control in mvASD. These results support the potential of sEMG as an objective tool for characterizing speech-motor performance and informing assessment approaches for minimally verbal individuals.