Objective: To examine the utility of speech biometrics for describing and delineating neurodegenerative diseases using acoustic measures and machine learning procedures.
Background: Progressive neurological disorders often result in changes to speech production. These changes can act as markers of disease onset or progression and treatment response. Acoustic analysis is an objective way to measure speech. Most work in this space has compared healthy controls with one disease, and not considered distinctions between multiple pathologies.
Method: We applied machine learning algorithms (support vector machines) to identify speech features to discriminate between different neurodegenerative diseases including multiple sclerosis (N=144) and Friederich’s ataxia (N=80), as well as healthy controls (N=174). Participants performed a diadochokinetic task where they repeated the alternating syllables /PA/, /TA/, and /KA/. Signal processing techniques were used to extract a wide range of spectral and temporal prosodic features from the speech recordings. Summary statistics of these acoustic features were subjected to machine learning.
Results: Data suggest that multiple-parameter acoustic sets distinguish neurodegenerative diseases including speech rate, pause and speech rate variability, spectral energy, spectral entropy, spectral crest, f0 duration, and spectral spread. Machine learning techniques produced high discrimination accuracy (M = ~84%) by simultaneously considering combinations of these acoustic features. Friederich’s ataxia (~82%) and multiple sclerosis (~91%) were both identified with high accuracy and sensitivity.
Conclusion: Speech biometrics can differentiate between neurodegenerative diseases and healthy speech with high accuracy, supporting the assumption of specific speech profiles in neurological disorders. We emphasize the importance examining multiple acoustic features when assessing key indicators of neurological disease.
To cite this abstract in AMA style:B. Schultz, Z.. Joukhader, U. Nattala, G. Noffs, J. Chan, S. Rojas-Azaocar, H. Reece, M. Magee, M. Delatycki, L. Corben, A. Walt, A. Vogel. Measurement of speech as a biomarker of neurodegenerative disease using acoustic profiles and machine learning [abstract]. Mov Disord. 2020; 35 (suppl 1). https://www.mdsabstracts.org/abstract/measurement-of-speech-as-a-biomarker-of-neurodegenerative-disease-using-acoustic-profiles-and-machine-learning/. Accessed December 1, 2023.
« Back to MDS Virtual Congress 2020
MDS Abstracts - https://www.mdsabstracts.org/abstract/measurement-of-speech-as-a-biomarker-of-neurodegenerative-disease-using-acoustic-profiles-and-machine-learning/