Voluntary cough detection by internal sound analysis

Cough can be defined as a forced expulsive onrush, normally against a closed glottis, producing a characteristic sound. It can be an indicator of many respiratory diseases, and its counting and classification is an important aspect. We propose a method on internal sound signal to automatically identify, count and (partly) qualify cough sounds. Our approach relies on explosive phase detection, because of its acoustic and spectral distinctive characteristics, and its potential for accurate onset detection of cough sounds. The features analyzed, related with tonality, pitch, timbre and frequency, prove to be very relevant in our explosive phase detection approach. Our results show a recall value of 86.6% and a precision value of 84.3%, for a wide testing population with and without respiratory perturbations. The internal sound analysis reveals advantageous in external noise reduction, therefore internal sounds are highlighted and better characterized.