Objective: This study’s objective is to validate functional vocal fold measurements obtained using flexible high-speed videoendoscopy (HSV) in the context of connected speech. The larger project is aiming at precision assessment that can optimize individual diagnostic and treatment outcomes in voice disorders. Custom-developed automated measures of glottal attack time (GAT), glottal offset time (GOT) and variation of left-right relative phase asymmetry (vA%) are validated against elicited hard, soft and habitual glottal attack in sustained phonation. The accuracy of the automated GAT, GOT and vA% measures are assessed relative to manually performed measures. Our previous studies suggested substantial differences in GAT and GOT between patients with adductor laryngeal dystonia and vocally healthy controls, as well as between soft and hard phonation. This study aims to provide the validity and accuracy assessment of GAT, GOT and vA% on a controlled set of HSV data before being implemented for analysis of connected speech.
Methods: HSV recordings at 5,000 fps of sustained phonations of the vowel /i/ of 40 vocally healthy subjects (20 men, 20 women) under the elicited hard, soft and habitual glottal attacks are analyzed. For each video recording, GAT, GOT and vA% are computed automatically and measured manually by a trained expert. The automated analysis consists of three steps: (1) temporal segmentation to identify the vocalized segments of the recording, (2) spatial segmentation of the vocal fold edges using a deep learning method, and (3) computation of the GAT, GOT, and vA% values.
Results and conclusion: The presentation will discuss: the validity of GAT, GOT and vA% measuring the differences between phonations with hard, soft and habitual glottal attacks; and the accuracy of the fully automated measures relative to the manually measured ground truth. Additionally, the results are compared to previous data from literature.
Acknowledgments: We acknowledge the support from NIH NIDCD R01DC019402 and K01DC017751.