To evaluate a generalized screening model integrating stress-induced acoustic analysis with machine learning for improved detection of respiratory disorders, particularly in the context of Post-Acute Sequelae of SARS-CoV-2 (PASC).
Approach:
Dataset: Utilized the DICOPERIA-Voice dataset (n = 154) for recordings of sustained vowel phonation (/a/) and voluntary coughing at resting state and after a physiological stress protocol.
Feature Extraction: Employed a dual-feature extraction strategy combining traditional acoustic biomarkers with high-dimensional Self-Supervised Learning (SSL) embeddings from wav2vec 2.0, WavLM, and HuBERT.
Classification: Performed binary classification (PASC vs. Healthy) using Logistic Regression, evaluated via stratified 5-fold cross-validation.
Key Findings:
Physical exertion significantly improved classification performance and reduced model variability across all tasks.
Fusion of acoustic features with WavLM and wav2vec 2.0 achieved peak F1-scores of 82.2% for vowel phonation and 80.8% for coughing in post-exercise conditions.
A cross-task late fusion model aggregation reached the highest overall performance with an F1-score of 87.7%.
Interpretation:
Incorporating Self-Supervised Learning representations into acoustic analysis improves the sensitivity of voice-based screening, while post-exercise measurements enhance robustness and consistency.
Limitations:
The study relies on a specific dataset which may limit generalizability.
Further validation is needed before integration into routine clinical assessments.
Conclusion:
The proposed framework offers a scalable and objective method for detecting respiratory and vocal sequelae in chronic or post-viral conditions.
by Álvaro Vera-López, Darío Tilves-Santiago, José Manuel Ramírez-Sánchez, Laura Docío-Fernández, Carmen García-Mateo, María Bustillo-Casado, Alejandro García-Caballero
Protection against spread appeared strongest within 6 months of vaccination, while exposed vaccinated contacts showed no measurable reduction in infection risk.
