Cough sound analysis and automated detection with vision transformers

This thesis described the design, implementation, and evaluation of a novel automated cough detection system based on audio recordings and machine learning techniques utilizing spectrogram analysis and vision transformer architectures. The primary data source for cough sound processing and data preparation were cough recordings from the RaDAR dataset, comprising 24-hour anonymized audio recordings collected from healthy volunteers and patients with various respiratory diagnoses. All recordings were obtained using the Vitalograph VitaloJAK cough monitoring device.