The Effect of Phonotrauma Severity on the Discriminative Power of Different Ambulatory Voice Measures

Objective: Phonotraumatic vocal hyperfunction (PVH) involves a “vicious cycle” where individuals compensate for pathology to sustain functional voice, but those compensations can exacerbate the phonotrauma. Within this framework, compensation increases with severity. We used a cross-sectional design to quantify severity effects on ambulatory voice measures such that severity-linked patterns have the potential to distinguish etiology from compensation in daily voice use. Method: Three laryngologists rated phonotrauma severity from still videostroboscopic images of 142 females with PVH (66 mild, 64 moderate, 12 severe). Daily vocal function and behavior were monitored for approximately 1 week in all patients and in 136 vocally healthy matched controls. Machine learning quantified the discriminative power of six ambulatory voice measures (difference between the magnitude of the first two harmonic magnitudes H1-H2, sound pressure level SPL, fundamental frequency f o , cepstral peak prominence CPP, voicing duration, and resting duration) across severity groups using logistic regression, nested 10-fold cross-validation, and wrapper-forward feature selection. Experiment 1 evaluated each measure separately for distinguishing mild from controls and moderate–severe from controls. Experiment 2 combined the significant measures from Experiment 1 to generate a composite classifier optimized for each severity level. Results: In the mild group, only H1-H2, SPL, and resting duration differentiated patients from controls. In the moderate–severe group, all measures except CPP were discriminative. H1-H2 showed the strongest severity dependence, with substantially higher discriminative power in the moderate–severe group, whereas SPL and resting duration performed similarly across severities. Composite models achieved average accuracies of 75.1% (mild), 79.9% (moderate), and 91.5% (severe). Conclusions: More severe phonotrauma was associated with increasingly atypical H1-H2 profiles reflecting more abrupt vocal fold closure, consistent with stronger compensatory patterns. In contrast, louder phonation and shorter vocal rests differentiated patients from controls regardless of severity, suggesting etiological contributions. These findings support treatment approaches that target both components: addressing loudness and rest patterns to modify etiologic tendencies, while simultaneously reducing abrupt vocal fold closure to address compensation. This dual focus may enhance prevention, intervention, and long-term outcomes for individuals with PVH.