14th International Congress of Phonetic Sciences (ICPhS-14) San Francisco, CA, USA August 1-7, 1999

Describing Hoarse Voices by Means of a Multivariate Analysis of Vowel Spectra

Fabrizio Bucella (3), Jean Schoentgen (1,2), Mounir Bensaid (1)

(1) Laboratory of Experimental Phonetics, Universit. Libre de Bruxelles, Brussels, Belgium
(2) National Fund for Scientific Research, Belgium
(3) Fonds pour la Formation à la Recherche dans l'Industrie et dans l'Agriculture, Belgium

The presentation concerns the description of hoarse voices by means of spectral cues obtained via a principal component analysis of contour-free spectra. Conventional spectral descriptors of hoarse voices involve the segregation of the harmonics from the rest of the spectrum. Possible problems are the following. First, the segregation of the harmonics is based on heuristic methods that make it possible to achieve an approximate separation only. Second, the spectral contour conditions the magnitudes of the harmonics and inter-harmonics and, therefore, their ratio that is involved in the spectral descriptors of hoarseness. We therefore propose describing spectra of hoarse voices by removing the spectral contour, ranking the spectral components according to their size, grouping components of similar size and performing a principal component analysis of the energies of the grouped spectral components. Results show that the first two principal components enable severely hoarse, moderately hoarse and clear voices to be distinguished.

Full Paper

Bibliographic reference.  Bucella, Fabrizio / Schoentgen, Jean / Bensaid, Mounir (1999): "Describing hoarse voices by means of a multivariate analysis of vowel spectra", In ICPhS-14, 2045-2048.