Objective: Creak—a low-frequency, irregular phonatory pattern—has emerged as a candidate acoustic biomarker for adductor laryngeal dystonia (AdLD). However, creak naturally occurs in American English through predictable linguistic mechanisms such as /t/ glottalization or final-phrase creak (also known as ‘vocal fry’). This study examined whether isolating creak that occurs outside these linguistically expected contexts enhances the acoustic distinction between AdLD, and typical voices compared to total (non-linguistic and linguistic) creak measurement.

Methods/Design: Speech samples were collected from 50 speakers with AdLD and 50 age- and sex-matched controls reading the first paragraph of the Rainbow Passage. Linguistically expected creak contexts were identified a priori using established phonological and prosodic processes. An automated algorithm detected creak percentage (%creak) within voiced segments. A linear mixed-effects model examined the interaction between group (AdLD, control) and creak type (linguistic, non-linguistic). Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of non-linguistic %creak versus total %creak.

Results: Although both groups used more linguistic than non-linguistic creak, the contrast between types was statistically significantly attenuated in speakers with AdLD. Non-linguistic %creak showed greater group separation, yielding a statistically higher diagnostic accuracy (AUC = .81) than total %creak (AUC = .74).

Conclusions: Pathological creak in AdLD disrupts the normal linguistic organization of creak, increasing its occurrence outside of linguistically expected contexts. Non-linguistic creak improves diagnostic discrimination compared to total creak, underscoring the value of linguistically informed acoustic biomarkers in improving diagnostic accuracy for AdLD.