| Abstract | Objective:
Vocal hyperfunction (VH) is associated with many common voice disorders and has been hypothesized to influence the baseline tension of the vocal folds during phonation. A measure of the relative fundamental frequency (RFF) during the onset and offset of phonation is thought to be related to this baseline tension. RFF has been used to differentiate between typical voices and those of patients with VH in laboratory settings using signals from both microphones and neck-surface accelerometer sensors. Similarly, it has been hypothesized that vocal effort increases with increase in tension of vocal folds. Hence, in this study, we investigated whether the laboratory sensitivity of RFF to vocal hyperfunction is preserved in naturalistic, in-field settings and how RFF can be used as a measure of vocal effort score.
Methods:
Ambulatory monitoring with accelerometer sensors placed on the anterior neck surface was conducted for both patients with vocal hyperfunction and vocally healthy control speakers in both laboratory and in-field, naturalistic settings. Automated RFF analysis on the ambulatory data was carried out, and the differences between the patient and control groups in both settings were analyzed. Machine learning classification methods were applied on RFF data to differentiate: 1) VH Patients and controls, 2) PVH patients and snowball sampled PVH controls, and 3) NPVH patients and controls. In order to investigate changes in vocal effort scores with the changes in RFF values, a linear regression classifier is trained using RFF data and self reported effort scores.
Results and Conclusions:
Results for RFF analysis for all subject groups in laboratory and naturalistic environments will be presented to answer whether the trends in RFF values observed for laboratory data generalize to naturalistic, in-field settings. Machine learning classifiers classified the VH patient group, PVH patient group and NPVH patient group with a maximum accuracy in the range of 72-77%. Preliminary results show limited prediction accuracy of self-reported effort scores based on RFF values. Overall, this study advances our understanding of using RFF as a biomarker for vocal hyperfunction in naturalistic environments, and classification algorithms presented here can not only be used as a monitoring tool but can also serve as a baseline for future RFF-based classification research.
|
|---|