A Gaussian mixture classifier model to differentiate respiratory symptoms using phonated /ɑː/ sounds

Balamurali B T, Hwan Ing Hee, Cindy Ming Ying Lin, Prachee Priyadarshinee, Christopher Johann Clarke, Dorien Herremans, Jer-Ming Chen

Poster presentation (paper to appear in proceedings)

Abstract

An audio-based classification model that differentiates between healthy vs pathological respiratory symptoms using acoustic features extracted from phonated /ɑː/ sounds is presented. For this, a new dataset of phonated /ɑː/ sounds, together with a clinician’s diagnosis, was compiled and a Gaussian Mixture Model (GMM) using Mel-Frequency Cepstral Coefficients (MFCCs) classifier was used. Despite no significant differences in mean values of the fundamental and formant frequency (F0, F1, F2, and F3) distribution for /ɑː/ sounds retrieved from healthy vs pathological populations, our /ɑː/ sound model trained using MFCCs resulted in an accuracy of 81.92% when compared against clinician’s diagnosis.