An interpretable, clinically grounded framework for digital speech biomarker development in neurodegenerative diseases.

Introduction: Communication ability-a key determinant of quality of life-is frequently affected and progressively declines in neurodegenerative diseases. Effective management of progressive communication disorders requires a personalized approach to deliver timely interventions tailored to the evolving profiles of communicative impairment, thereby supporting functional communication throughout the disease course. To this end, reliable tools capable of detecting and quantifying both disease-specific patterns of communicative impairment and within-disease phenotypic variability are urgently needed. This study leverages Artificial Intelligence and advanced data analytics to develop an acoustic-based framework for automated extraction of interpretable, clinically grounded speech markers to enable objective assessment and phenotyping of progressive communication disorders. Methods: Three groups of participants, including 14 individuals with amyotrophic lateral sclerosis (ALS) and 15 individuals with Parkinson's disease (PD), alongside 10 neurologically healthy controls, performed a standardized oral passage reading task, yielding 739 speech samples. Fifty acoustic features were extracted using an automated analytic pipeline and subsequently clustered into six interpretable composite markers. The clinical utility of these markers was evaluated with the recorded speech samples by examining their (1) associations with standardized metrics of cognitive, motor speech, and overall communicative functions, (2) efficacy for detecting and differentiating disease-specific communicative impairment patterns in ALS and PD using supervised machine learning, and (3) utility for within-disease phenotyping and stratification using unsupervised clustering analysis. Results: The markers effectively (1) detected subtle subclinical changes across multiple domains prior to substantial declines in functional communication outcomes; (2) differentiated disease-specific patterns of communicative impairment (multiclass area under the curve > 0.90); and (3) identified subgroups with distinct speech profiles within each disease. Discussion: The findings support the potential of the proposed framework as a clinically translatable, objective tool to facilitate early detection, differential diagnosis, and phenotyping of progressive communication disorders, ultimately advancing personalized, measurement-based care in neurodegenerative diseases.