The flexibility of dynamical systems has recently been described as being insufficient to account for perturbated or highly noisy speech (Parrell & Lammert 2019), such as the speech output of patients with movement disorders (Parkinson’s disease). The shape of the kinematic contours can deviate considerably from those predicted by the models and an application of these models to impaired speakers can be even more difficult. In a critical discussion, I propose the introduction of stochastic noise as a random variable in dynamic simulations. I will conclude that noise plays also a role in modelling neurotypical speech – ranging from syllable repetition tasks to natural sentence production.