Toward an Articulatory Weakness Index for Speech Kinematics in Parkinson's Disease

Many acoustic speech metrics often reflect a combination of respiratory, laryngeal, and articulatory contributions, thereby preventing direct measurement of articulatory dynamics. We have developed an articulatory weakness index (AWI) derived from electromagnetic articulography (EMA) or acoustic data. We learn an articulatory-state coordinate $Z_a$ from healthy EMA data, where higher values indicate restricted articulator motion. In cohorts of individuals with Parkinson's disease (PD) and controls, AWI is elevated in PD and correlates with motor severity, while remaining stable across in healthy controls and speakers with vocal hyperfunction. AWI provides an interpretable, disentangled articulatory weakness index from acoustic speech recordings, complementing respiratory and laryngeal measures.