Automated Detection of Vocalizations in Laryngeal Videostroboscopy.

The glottal area (GA) analysis during vocal fold vibration has garnered significant attention in phonatory medicine. Clinicians can obtain quantitative vocal fold characteristics that enhance diagnostic accuracy by examining endoscopic images captured during phonation. We propose an automated computational system for videostroboscopy that leverages spectrograms and cosine similarity techniques. This system efficiently identifies vocalization areas that warrant further analysis. Our Voice Activity Detection (VAD) method demonstrates over 97% precision and recall based on 5-fold cross-validation. In clinical applications, this technique will allow healthcare providers to visualize specific regions of vocal fold vibration more rapidly and correctly, leading to better diagnostic results.