Validating automated speech timing methods in clinical and healthy speakers across sentence, paragraph, and monologue tasks

Automated measurement of speaking and articulation rates holds promise as a scalable alternative to manual analysis in clinical populations. This study evaluated a Praat-based script that estimates global speech timing by detecting syllable nuclei via amplitude dips. Speaking rate (syllables/total duration) and articulation rate (syllables/speaking time) were measured manually and with an automated script across speakers with multiple sclerosis (MS), Parkinson's disease (PD), and healthy controls. Sixty participants (20 per group) completed sentence, paragraph, and monologue tasks (N = 180 recordings). Default script parameters were compared to an optimized version with manually tuned dip thresholds. Analyses included error metrics, linear mixed-effects models, and generalizability analysis. Automated speaking rate measures showed strong correlations with manual measures across all groups and tasks (r = 0.623–0.998). However, default automated estimates underestimated both speaking and articulation rates, especially in clinical speakers and for the monologue task. Articulation rate was more sensitive to the measurement method, which accounted for nearly half of the total variance. Optimization of the Praat script parameters reduced proportional error by ∼60%, with varying effects across groups. Findings suggest that optimized automated methods can improve measurement accuracy, but population- and task-specific challenges persist, especially for articulation rate in MS and PD speakers.