Researchers use AI speech analysis to predict ALS speech impairment at 86% accuracy

Amyotrophic Lateral Sclerosis (ALS) often impairs speech due to bulbar dysfunction, but early detection can improve quality of life. In a study accepted at Interspeech 2026 in Sydney, researchers predict speech impairment in 66 German-speaking individuals with ALS using two clinical speech-related scores. They evaluated cross-sectional (across speakers) and personalized (within-speaker) modeling paradigms. Common speech tasks, particularly repetition tasks (/da/-/da/, /da/-/ba/), were analyzed for their utility. The cross-sectional model achieved a Concordance Correlation Coefficient (CCC) of 0.62 for predicting the Quality of Life in the Dysarthric Speaker questionnaire, while the personalized within-speaker setting reached a CCC of 0.86.

The study highlights the potential of automated speech analysis as a supportive tool for speech impairment assessment in ALS. By standardizing data collection and leveraging simple repetition tasks, the approach could enable non-invasive monitoring of disease progression. Future work may expand to larger cohorts and other languages. The paper, authored by Monica Gonzalez-Machorro, Ricarda von Heynitz, and colleagues, represents an initial step toward integrating AI into clinical workflows for ALS care.