Abstract

Assessment of prosody is important for diagnosis and remediation of speech and language disorders, for foreign language instruction, and for diagnosis of neurological conditions, including Autism Spectrum Disorder (ASD). Current assessment methods are largely auditory-perceptual. In this talk, I propose an alternative approach for evaluating focus and lexical stress automatically using speech technology. Speech data were collected from children with ASD and with typical development (TD) who performed three tasks, each designed to elicit a contrast in stress or focus. Our automated analysis of these data combined global spectral, fundamental frequency, and temporal information into a single score. These automatically computed scores were validated against scores both from naïve judges and trained clinicians. We found that these automated measures correlated with the scores produced by the judges better than the clinicians’ scores did and as well as the individual judges correlated with one another. The automatically generated scores also differentiated significantly between the ASD and TD groups but revealed that the acoustic features that predict auditory-perceptual judgment are not the same as those that distinguish the two diagnostic groups.