Can speech-based measures support Developmental Language Disorder identification? An explorative study.

This paper aims to identify possible phonetic biomarkers of Developmental Language Disorder (DLD) in Italian preschool children. Speech samples, collected during three retelling tasks, were transcribed and processed through a computational pipeline. A set of acoustic and rhythmic parameters were automatically extracted from the recordings and analysed by using descriptive and inferential statistics. Our work demonstrates that i) language difficulties of DLD take the form of reduced fluency and speech disruptions ii) some acoustical characteristics of the voice (e.g., the mean value of the fundamental frequency) can distinguish language-impaired children from peers. These results suggest that automatic voice and speech analysis could provide new markers of the DLD - markers that are not audible to the human ear and therefore fall outside the possibilities of conventional paper-and-pencil neuropsychological tests.