Quantification of Speech Disfluency as a Marker of Medication-Induced Cognitive Impairment: An Application of...

We present the results of a study investigating the use of speech and language characteristics extracted from spontaneous spoken discourse to assess changes in cognitive function. Specifically, we investigated the use of automatic speech recognition technology to characterize spontaneous speech disfluency induced by topiramate, an anti-epileptic medication with language-related side-effects. We audio recorded spontaneous speech samples from 20 participants during several picture description tasks and analyzed the recordings automatically and manually to extract a range of spoken fluency measurements including speech discontinuities (e.g., filled pauses, false starts, and repetitions), silent pause duration, speaking rate and vowel lengthening. Our results indicate that some of these paralinguistic speech characteristics are a) sensitive to the effects of topiramate, b) are associated with topiramate concentrations in the blood, and c) complement standard neuropsychological tests typically used to investigate cognitive effects of medications. This work demonstrates the use of computational linguistic tools to assess cognitive effects in a more sensitive, objective and reproducible manner than is currently available with standard tests.