Longitudinal changes of automated speech measures in natural connected speech in ALS

Abstract

AbstractBackgroundAmyotrophic Lateral Sclerosis (ALS) is a multi‐system neurodegenerative disorder with motor and cognitive deficits. Speech is affected both by motor processes associated with the articulation of utterances and by cognitive processes related to conveying a linguistic message. Previous studies in ALS have reported reduced speaking rate, longer pause durations, and more pause events compared to healthy controls. However, existing studies are scarce, rarely compare the effects of motor versus cognitive deficits, and have not yet characterized longitudinal changes in this population. In this study, we analyzed longitudinal changes in speech acoustic features in ALS from 1‐minute speech samples using an objective, reproducible, fully automated method.MethodWe examined digitized Cookie Theft picture descriptions produced by 21 native English‐speakers (66.7% males, mean (SD) age (y) = 65.3 (11.2); education (y)= 15.2 (2.6)) with ALS for up to 6 years (mean inter‐sample interval= 12.8 months, median samples/subject = 3). We automatically segmented each sample into speech and pause segments and extracted total speech = sum of speech segment durations (sec); total time = sum of speech and pause segment durations (sec); average speech and pause segment durations (sec); and speech frequency and fragment rate = number of speech events per 60 sec of total time and total speech, respectively. Longitudinal changes were related to cognition (ALS‐Specific Edinburgh Cognitive and Behavioral ALS Screen scores (ALS‐sp. ECAS)), motor features (ALS Functional Rating Scale–Revised scores (ALSFRS‐R)), and progression rate (change in ALSFRS‐R over disease duration) using linear mixed‐effects models with interactions between time and the predictor of interest.ResultLower scores on ALS‐sp. ECAS were predictive of an increase in pause durations (p=.023) and a reduction in total time (p = .019) over time. Higher ALSFRS‐R scores were predictive of increased speech fragment rate (p=.013) over time. Faster progression rates were associated with longer pauses (p=.006) and reduced speech frequency (p=.001) over time.ConclusionSpeech acoustic measures are differentially impacted by motor versus cognitive features over the disease course in ALS. Longitudinal speech changes in ALS can be characterized using automated methods of acoustic analysis natural speech.