Trajectories in speech production: can optimal control markedly shape the intrinsic biomechanical dynamics?

International audienceFrom the previous work of one of us, we know that the biomechanics of the articulators have a decisive influence on the shape of speech trajectories. Recently, we examined whether, on top of that, an optimality principle can be useful to account for trajectories from schwa to vowel production. Here we wish to push further our analysis and study how it can affect trajectory formation in longer phoneme sequences. In order to achieve that, we implemented an optimal control model for speech production. The planner, which assumesthat the central nervous system is stocking association between values of the control variables corresponding to goal positions and the corresponding formant frequencies, and the plant to be controlled, which is a biomechanical model of the tongue in our case, were taken from the GEPPETO model of speech motor control. In the optimal control model, trajectories were determined by the selection of motor command that jointly minimized the motor effort and the motor error (bias in production). The internal model used for that optimization was a LSTM trained from the biomechanics of GEPPETO. Thus trajectories reflected both a representation of biomechanics and a higher-order effort constraint. In contrast, in the GEPPETO model, trajectories emerged from the actual biomechanical properties of the articulators and the only specification of sensory targets. We contrasted the two models in [VV] and [VCV] sequences and compared the results obtained in either the acoustic, kinematic, or motor domain