Segment-Based Signal Typing and Predictive Modeling in Pediatric Dysphonia With Different Vibratory Sources

Area of Science:

• Speech-language pathology
• Acoustic analysis
• Voice disorders

Background:

• Severe dysphonia in children presents challenges for acoustic analysis due to signal aperiodicity.
• Current voice signal typing methods are subjective and may not capture multiple signal types within a sample.

Purpose of the Study:

• To refine a manual signal typing tool for segment-level labeling of pediatric voice signals.
• To develop an objective, predictive model for automated voice signal typing in children.

Main Methods:

• Expert speech-language pathologists manually labeled voice samples from children with glottal and supraglottal vibratory sources.
• A predictive model was trained using acoustic measures like pitch strength, EnvSD8, sharpness, and CPPS.
• Model performance was assessed using a test set and cross-validation.

Main Results:

• Manual typing identified multiple signal types in 11% of samples overall and 20% of samples with supraglottal vibratory sources (SGVS).
• The predictive model achieved 81%-96% accuracy in classifying signal types.
• Key acoustic measures (EnvSD8, CPPS, sharpness) were effective predictors.

Conclusions:

• The refined manual tool enhances signal typing precision, potentially improving acoustic measure reliability and enabling new outcome metrics.
• Automated signal typing using objective measures offers significant clinical utility for pediatric voice analysis.