Improving respiratory disease detection through SSL-enhanced acoustic analysis and exercise-rest measurements

The study examined the possibility of detecting respiratory diseases through voice and cough analysis combined with modern machine learning. The research included 154 patients who were recorded at rest and after physical exertion (six-minute walk and sit-to-stand test). Researchers combined traditional acoustic features with self-supervised learning representations (wav2vec 2.0, WavLM, and HuBERT). Physical exertion significantly improved diagnostic accuracy. The best results were achieved by combining all methods with an F1-score of 87.7 percent. This method could serve as a rapid and non-invasive tool for detecting consequences of respiratory diseases, including long-term effects of COVID-19.