Related Experiment Video
Updated: May 30, 2026
12:43
A Protocol for Comprehensive Assessment of Bulbar Dysfunction in Amyotrophic Lateral Sclerosis ALS
Published on: February 21, 2011
35.0K
Using Digital Speech Markers to Classify Functional Speech Disorder: A Proof-of-Concept Pilot Study.
Jennifer L Freeburn1,2, Sara A Finkelstein2, Christiana Westlin2,3
1Department of Speech, Language and Swallowing Disorders, Massachusetts General Hospital, Mass General Brigham Integrated Healthcare System, Boston, Massachusetts, USA.
Summary
Quantitative speech analysis can help diagnose functional speech disorder (FND-speech). This study found acoustic and linguistic speech features effectively distinguish FND-speech from healthy controls, showing potential as diagnostic markers.
Area of Science:
- Neurology
- Speech Science
- Computational Linguistics
Background:
- Functional speech disorder (FND-speech) is a subtype of functional neurological disorder.
- Quantitative characterization of FND-speech motor and cognitive-linguistic features is underexplored.
Purpose of the Study:
- To quantitatively characterize FND-speech by comparing acoustic and linguistic features.
- To evaluate digital speech features as potential adjunctive diagnostic markers for FND-speech.
Main Methods:
- Compared 30 adults with FND-speech and 47 healthy controls (HCs).
- Utilized lexicosyntactic, rate-based, and acoustic markers from a structured picture description task.
- Employed supervised machine learning for FND-speech versus HC classification.
Main Results:
- Lexicosyntactic features showed moderate predictive power (AUC=0.80).
- Rate-based features demonstrated moderate predictive power (AUC=0.81).
- Acoustic features achieved high discrimination (AUC=0.98), with a combined model yielding similar performance (AUC=0.98).
Conclusions:
- Successfully classified FND-speech versus HCs using digital speech markers.
- Highlights the potential of speech markers as adjunctive diagnostic tools for FND-speech.
- Further validation with out-of-sample replication and larger datasets including neurological controls is recommended.
