Instrumental dimensioning of normal and pathological phonation using acoustic measurements

Abstract

The present study deals with the dimensions of normal and pathological phonation. Separation of normal voices from pathological voices is tested under different aspects. Using a new parametrization of voice-quality properties in the acoustic signal, the vowel productions of 534 speakers (267 M, 267 F) without any reported voice pathology and the productions of 534 gender-matched pathological speakers were considered. In a first step, a gender-specific separation of the two groups is supported by a number of significantly different parameter means. In a second step, a clustering technique differentiates three subgroups within each group and gender on the basis of the acoustic parameters. Further, a statistical examination of the correct assignment in the database (DB-classification) as "normal" or "pathological" shows that the two groups overlap to some extent. The overlap of speaker assignment indicates a phonation continuum through the multidimensional space extending from normal to pathological voices. The validity of a categorical distinction of normal and pathological phonation in the sense of an individual or group-orientated labelling of voice quality as "normal" or "pathological", respectively, is discussed.