Characterizing Vocal Hyperfunction Using Ecological Momentary Assessment of Relative Fundamental Frequency.

Many common voice disorders are associated with vocal hyperfunction (VH), with subtypes including phonotraumatic VH (leading to organic vocal fold lesions such as nodules and/or polyps) and nonphonotraumatic VH (often diagnosed as primary muscle tension dysphonia). VH has been hypothesized to influence baseline vocal fold tension during phonation, and the relative fundamental frequency (RFF) during onset and offset cycles of phonation has been related to vocal fold tension and has been shown to differentiate typical voices from patients with VH in laboratory settings. In this study, we investigated whether the laboratory sensitivity of RFF to the presence of VH found in the laboratory is preserved in naturalistic, in-field settings and whether ecological momentary assessment of RFF during daily life could be a correlate of self-reported vocal effort. RFF analysis was carried out after performing smartphone-based monitoring of anterior neck-surface vibration with accelerometer sensors in both laboratory and in-field settings. Supervised machine learning was applied to combine multiple RFF values to discriminate and classify patients with VH from vocally typical speakers. Results showed that RFF-based classification of VH can be preserved in the naturalistic environments for patients with phonotraumatic (81.3% accuracy) and nonphonotraumatic (62.5% accuracy) VH. Additionally, we used explainability techniques to understand which RFF features were clinically relevant in the classification tasks. No direct relationship was observed between RFF and self-reported vocal effort. Overall, this study advances our understanding about RFF as a potential biomarker of VH as individuals go about their daily life. Machine learning algorithms can be implemented within a monitoring device for proactive screening or in biofeedback-based voice therapy paradigms.