Acoustic and linguistic features of reading reveal early change, progression and function in ataxias

A major obstacle for clinical trials is the lack of objective, sensitive, and reliable measures that can detect modest changes in disease progression. Here, we determine whether acoustic and linguistic digital speech measures automatically obtained during a functionally relevant passage-reading task capture multiple dimensions of disease in ataxia, including functional communication impairment, subclinical cerebellar dysfunction and disease progression. A total of 157 individuals with ataxia and 84 controls contributed cross-sectional data, and 54 individuals with ataxia and 43 controls contributed longitudinal data within the ongoing Neurobooth natural history study. Participants completed standardized speech recordings, patient-reported outcome measures (PROMs) and neurologist-rated clinical evaluations. A novel speech processing pipeline was developed to automatically transcribe audio recordings, identify word boundaries and extract a predefined set of linguistic and within-word acoustic features. Individuals with ataxia exhibited marked disruption of speech timing, coordination and articulatory control, including slowed speech (d=1.23), prolonged inter-word pauses (d=-0.91), higher/more variable vocal intensity (|d|=0.43-0.51) and altered spectral content (|d|=0.43-0.79) compared to healthy controls. Linguistic features (e.g. speaking rate and within-word pause duration) showed strong associations with clinician-rated severity and PROMs (|r|=0.23-68), indicating alignment with functional communication impairment and patient-perceived disease burden. In contrast, acoustic features derived from cepstral measures captured subtle abnormalities in speech motor control, differentiating not only individuals with ataxia (d=0.65) but also pre-ataxic individuals (d=0.56), and those without clinically evident dysarthria (d=0.45), from controls. These findings indicate that acoustic features reflect subclinical cerebellar motor dysfunction involving impaired temporal coordination and vocal control before overt clinical speech impairment emerges. Longitudinally, several acoustic measures were sensitive to disease progression (MSDR=0.19-0.68), even in cases where clinical scales showed no detectable change. Speech-derived changes correlated with changes in clinical scales and PROMs. Both acoustic and linguistic features exhibited strong intra-session reliability. During passage reading, acoustic and linguistic measures provide complementary but different clinical information in ataxias. Linguistic measures primarily reflect downstream functional consequences of ataxic dysarthria, whereas acoustic measures provide sensitive indicators of subclinical cerebellar motor dysfunction and progression. These findings demonstrate that natural speech analysis can produce digital measures for detecting subclinical disease, quantifying functional impairment, monitoring progression in ataxia, with strong potential for application in clinical trials and remote monitoring.