Speech as a digital biomarker in multiple sclerosis: Automatic analysis of speech metrics using a multi-speech-task protocol in a cross-sectional MS cohort study

Abstract

Background:

Abnormalities in speech are well-documented in people with multiple sclerosis (pwMS). Recent studies focused mainly on acoustic measures, limiting the range of speech features analyzed. The aim of this study is to include both acoustic and linguistic features and evaluate their relationship with cardinal symptoms of multiple sclerosis (MS).

Methods:

In a prospective study, speech of N = 282 native German speakers (145 healthy controls (HCs) and 137 pwMS) was recorded at the University Clinic Dresden. Participants completed an extensive speech protocol, comprising narrative, articulatory, and phonatory tasks. Speech was analyzed using the SIGMA speech features library.

Results:

Differences in pwMS compared with HCs were evident in instabilities in phonation and articulation (i.e. jitter: H = 8.17, p = 0.04, d = 0.33; standard deviation of loudness: H = 14.17, p = 0.01, d = 0.45) during narrative speech tasks with high cognitive load. Speech abnormalities correlated with the Expanded Disability Status Scale (i.e. jitter: r = 0.25, p < 0.01; loudness peaks: r = 0.26, p < 0.01).

Conclusion:

Phonatory instabilities, changes in pitch and loudness, and reduced speech time are associated with disease severity, motor and cognitive symptoms. Speech parameters related to motor functions are influenced by cognitively demanding tasks, suggesting an interaction between functions.