In view of these observations, it seems likely that the data-based model of the vowel dynamics possesses an important degree of structure, perhaps reflecting physiological considerations, that requires further investigation. It is also clear that whilst encouraging there is still some way to go in overcoming the limitations of the approach. It is clear that speech is a nonlinear process and that if we are to achieve the holy grail of truly natural sounding synthetic speech than this must be accounted for. It is also clear that nonlinear synthesis techniques offer some potential to achieve this although a great deal of research work remains to be done. In the field of speech recognition, there is also strong experimental evidence that acoustic features representing various aspects of the nonlinear structure of speech can increase the robustness of recognition systems. However, more research is needed to find optimal ways for fusing the nonlinear with the linear speech features.