Laryngeal Pathologies Analysis Using Glottal Source Features

This experimental study looks at the problem of pathological disorders analysis from speech signals. Knowing that the vocal tract pathologies provide acoustic changes commonly called dysphonia. In particular, laryngeal pathologies have an effect on the larynx and vocal cords. When regarding to the speech production model, source-filter, the impact is even more apparent in the glottal source. Our study reports on : firstly, the estimation of the glottal source where the accurate estimation is considered. Secondly, parameterization of the glottal source. Many features are extracted such as normalized amplitude quotient (NAQ), difference between harmonic H1H2, peak slop . This analysis is performed around the German database which contains many recordings (normal and pathological subjects). More than 50 subjects are used where each speaker pronounce the sustained vowel [a] in normal intonation. The extracted parameters show an interesting tool for the discrimination between normal and pathological voice. In particular, the peak slop and harmonic richness factor present a significant differences. This work is developed under Matlab thanks to the project COVAREP (A Collaborative Voice Analysis Repository for speech technologies).