15th International Congress of Phonetic Sciences (ICPhS-15) Barcelona, Spain August 3-9, 2003

Speaker Normalization Using an Articulatory Model

Arturo Galvan (1), Rafael Laboissière (2)

(1) ITESM, Mexico
(2) Max Planck Institute for Psychological Research, Germany

We present a methodology for normalizing vowel formant values. Such methodology addresses the issues of vocal tract length compensation as well as articulatory strategy. Using an articulatory model of the vocal tract, we first estimate vocal tract dimensions in order to obtain a version of the model capable of covering the formant space of a given speaker. We then compel such model to produce the subject's observed vowel formants. We finally approach the articulatory parameters of all speakers to a set of per-vowel canonical articulatory parameters.

Full Paper

Bibliographic reference.  Galvan, Arturo / Laboissière, Rafael (2003): "Speaker normalization using an articulatory model", In ICPhS-15, 1325-1328.