Lexical and acoustic speech features relating to Alzheimer’s disease pathology
AbstractINTRODUCTIONIn this study, we compared digital speech features of AD and lvPPA patients in a biologically confirmed cohort and related them to specific neuropsychiatric test scores and CSF proteins.METHODSWe extracted language variables with automated lexical and acoustic pipelines from oral picture descriptions of 44 AD and 21 lvPPA patients with autopsy or CSF confirmation of AD pathology. We correlated distinct speech features with MMSE and BNT test scores and CSF p-tau levels.RESULTSLvPPA patients produced fewer verbs, adjectives, and more fillers with lower lexical diversity and higher pause rate than AD. Both groups showed some shared language impairments compared with normal speakers.DISCUSSIONOur speech measures captured differences in speech between the two phenotypes. Also, shared speech markers were linked to the common underlying pathology. This work demonstrates the potential of natural speech analysis in detecting underlying AD pathology.