This paper presents a procedure for normalization based on the exploitation of the concept of formant to cavity affiliation as described in the vocalic theory of speech production. For closed vowels, this concept allows an estimation of the ratio of the respective back and front cavity lengths of two speakers, from the ratio of the associated formants. In order to propose a normalization of the vowel space of different speakers we used a "reference speaker", whose articulatory and acoustic properties are well defined: the articulatory model as proposed by Maeda. Vowel normalization consists then in the projection of the vocalic space of one given speaker into the vocalic space of our model. Geometrical relations between the studied speaker and the "reference speaker" are extracted from [i] and [u] for front and velo-palatal vowels, and from [a] and [u] for back and velo-pharyngeal vowels, [y] and [u] give some indications about lip shapes. This method is tested (1) by its ability to predict the standard vowels of a speaker starting from those of the "reference speaker", and (2) by its ability to propose realistic formant patterns for Maeda's model, starting from the corresponding formant transitions of a real speaker. Keywords : speaker normalization; speaker adaptation; vowel production.
Bibliographic reference. Payan, Yohan / Perrier, Pascal (1993): "Vowel normalization by articulatory normalization first attemps for vowel transitions", In EUROSPEECH'93, 417-420.