Application of a roughness model to pathological voices

Irregularities in voiced speech are often related to the intrinsic nonlinearities in the vibrations of the vocal folds. Desynchronization of vibratory modes can induce nonlinear phenomena such as subharmonics, biphonation (two independent pitches), and deterministic chaos. The resulting complex signals exhibit various amplitude and frequency modulations and are, therefore, perceived typically as rough voices. In clinical studies of pathological voices perceptive roughness scores are widely used. So far, however, no satisfactory correlation of ratings of roughness and acoustic parameters has been found. In this presentation a modified Aures model for the quantification of roughness is introduced. It is exemplified that subharmonics lead to high roughness scores whereas the roughness of biphonation depends strongly on the two frequencies. Finally, a set of 120 perceptually evaluated pathological voices is studied. It turns out that the roughness score obtained from the psychoacoustic model is stronger correlated to the ratings of phoniatricians than conventional perturbation measures such as jitter and shimmer.