A Hybrid Machine-Learning-Based Method for Analytic Representation of the Vocal Fold Edges during Connected Speech.

Area of Science:

• Speech science and technology
• Biomedical engineering
• Medical imaging analysis

Background:

• Accurate detection of vocal fold edges during vibration is crucial for understanding phonatory processes in connected speech.
• High-speed videoendoscopy (HSV) provides detailed visual data of vocal fold dynamics.
• Automated segmentation of vocal fold edges in HSV data remains a challenge.

Purpose of the Study:

• To develop and validate a novel spatio-temporal technique for automatic vocal fold edge segmentation in HSV data during running speech.
• To create a robust and computationally efficient automated tool for analyzing vocal fold vibratory features.

Main Methods:

• A hybrid approach combining unsupervised machine learning (ML) and active contour modeling (ACM) was developed.
• K-means clustering was used to identify the glottal area from HSV kymograms, initializing the ACM.
• The ACM algorithm precisely detected the glottal edges of vibrating vocal folds across different cross-sections.

Main Results:

• The developed hybrid algorithm accurately segmented and tracked vocal fold edges in HSV data during connected speech.
• The method demonstrated high robustness against image noise.
• The technique achieved low computational cost, enabling efficient analysis.

Conclusions:

• The proposed hybrid ML-ACM technique offers a fully automated and accurate solution for vocal fold edge detection in HSV.
• This tool facilitates detailed analysis of vocal fold vibratory features in connected speech.
• The findings contribute to advancements in speech production research and clinical diagnostics.