Improved articulatory models

Abstract

After a brief historical review of articulatory models, a factor analysis of the lateral shapes of the vocal tract is described. In the analysis, the profiles are specified by the sum of a small number of linear components. Studies at CNET indicate that, by using an arbitrary factor analysis in connection with the principal component analysis, it is possible to derive a linear model in which each component can be interpreted in articulatory terms, such as jaw position, tongue-body position, etc. The statistical model, therefore, can be regarded as an articulatory model. The analysis of radiocinematographic data exhibits that the same vowel can be produced differently depending on speakers and phonetic contents, suggesting a compensatory articulation. Such a model is also useful in investigating the articulatory-acoustic relationships. For example, F1–F2 trajectories are calculated by varying a specific parameter value, while the remaining parameter values are fixed appropriate for a vowel. The results show that the different jaw positions can be compensated only by the tongue-body position for front vowels, and only by the lip aperture for back vowels, to obtain a similar F1–F2 pattern. Finally, limitations of current articulatory models are discussed.