Establishing some principles of human speech production through two-dimensional computational models

Human speech production is often described as an optimisation process, which tends to maximise the effectiveness of the communication process minimising the effort involved in the production. The aim of this paper is to investigate this highly complex problem with two dimensionally reduced spaces corresponding to different computational models. Since the highdimensional parameter space which usually describes such a problem is often an issue in the optimal-behaviour computation, two-dimensional models are proposed. The first one analyses the best trajectories visiting the proximity of a set of randomly chosen points. The second one explores the F1-F2 vowel space trying to maximise a set of likelihood functions describing some human production characteristics. Even though such models need further development, some preliminary correspondences can be observed with some of the elements described in the most popular theories for human speech production. For example, the distance between close competitors directly influences the best trajectory computation and, therefore, the effort needed to achieve the desired tasks. The trajectory planning is also controlled by the degree of motivation selected to achieve the desired accuracy: the higher the motivation, the more the target must be addressed.