From Screening to Precision: Searching for Voice Disorder-Specific Acoustic and Auditory-Perceptual Metrics.

BACKGROUND

Acoustic and auditory-perceptual parameters are common tools for screening clinically significant voice disorders. However, the potential for disorder-specific acoustic signatures that support clinical differential diagnosis remains largely unrealized. Additionally, the robustness of acoustic patterns across different speech materials requires clarification to inform flexible, evidence-based clinical protocols and emerging machine learning applications.

METHODS

This study investigated disorder-specific metrics and speech material consistency using the Perceptual Voice Qualities Database. Generalized Linear Models examined associations between 14 acoustic parameters and common voice pathologies [Vocal Fold Paralysis (VFP), Atrophy, Lesions, and Muscle Tension Dysphonia (MTD)]. Principal component analysis (PCA) integrated acoustic and auditory-perceptual measures to identify multidimensional voice quality patterns, while Receiver Operating Characteristic (ROC) curves evaluated discriminative performance across sustained vowels and connected speech.

RESULTS

Two primary principal components emerged: PC1 (34.7% variance) integrating general voice quality and perceptual ratings, and PC2 (17.3% variance) contrasting temporal stability with harmonic structure. Distinct disorder-specific patterns were identified: VFP demonstrated strong discriminative performance on both components (AUC ≥ 0.75), while Atrophy, Lesions, and MTD showed moderate associations with PC1 (AUC = 0.52-0.66). Preliminary analysis revealed characteristic patterns for Parkinson's disease across both components. Importantly, acoustic patterns remained consistent across speech materials, supporting task-flexible clinical assessment protocols.

CONCLUSION

Specific voice pathologies exhibit distinct acoustic-perceptual signatures that can be reliably identified through multidimensional analysis. These findings support a precision-based approach to voice assessment, moving beyond general screening toward disorder-specific diagnostic applications. The robustness of patterns across speech materials enables flexible clinical protocols, while the integration of acoustic and perceptual measures provides a foundation for enhanced diagnostic tools and machine learning applications.