The “sequential neural network” proposed by Jordan (MIT COINS Technical Report 88-27, 1988) enables learning of motor skill problems involving excess degrees of freedom. This structure was used with (i) the articulatory model developed by Meada [J. Acoust. Soc. Am. Suppl. 1 65, S22 (1979)] as an internal “forward” model relating a given set of articulatory commands (five articulators, e.g., lips, jaws, tongue body, tongue dorsum, and tongue tip) to their acoustic consequences, (ii) vocalic targets specified in the formant (F1, F2, F3) space, and (iii) smoothness constraints leading to coarticulation phenomena. A first set of results about learning of such vocalic gestures will be presented and the following will be discussed: (i) the pattern of coarticulation phenomena observed in this structure, such as between-articulator relationships for achieving a given formant target and temporal organization of the trajectories in the articulatory space, (ii) the role of gesture duration, and (iii) the compensation ability of the network in relation to perturbations such as those observed in the bite-block experiments by Lindblom et al. [J. Phonet. 7, 147–162 (1979)].

The “effective second formant” F′2 was analytically tested as a cue for rounding opposition in French front vowels (/i/-/y/ and /e/-/ø/ opposition). For this purpose, a corpus of these four vowels introduced into contexts maximizing and minimizing the rounding opposition [C. Abry, L-J. Boë, P. Corsi, R. Descout, M. Gentil, and P. Graillot, Labialité et Phonétique (Université des Langues et des Lettres, Grenoble, France, 1980)] was acoustically studied. The first four formant frequencies and levels were measured. The F′2 of the analyzed vowels was next estimated with a model based on psychophysical and vowel perception hypotheses and using the frequencies and levels of F2, F3, and F4 [M. Mantakas, J-L. Schwartz, and P. Escudier, Actes des 15e Journées d'études sur la Parole (GALF, Aix-en-Provence, France, 1986), pp. 157–161]. The effectiveness of F′2 as a classification parameter was furthermore compared with that of other acoustic cues proposed for French. [Work supported by CNET, France.]

“Gross” vowel spectrum parameters for vowel classification are of interest to many researchers. In the present study, the “effective formant” F′2, estimated by the large-band spectral integration (LBI) model [Escudier et al., Acts of the French-Swedish Seminar, Grenoble, France (1985)], in the classification of natural French front vowels /i/ vs /y/ and /e/ vs /ø/(rounding opposition), is evaluated. In parallel, formants' measures on the same corpus are provided in order to compare with and interpret the results obtained using the LBI model Classification performance is good in the case of /i/-/y/. The optimal spectral integration window is 2–2.5 Bark large (cf. with the 3–3.5 Bark “critical distance”). The model fails, however, in the /e/-/ø/ classification. Here, the second formant's frequency is by far winning. The model's weakness apparently resides in its last stage, namely peak estimation. In fact, recent results show that the LBI spectral representation can support the definition of form factors (other than peak position) for the classification of vowels.