Computational Exploration of Pathological Voice Source via Global Nonlinear Dynamical Modeling

ABSTRACT Existing vocal folds models have limitations in simulating abnormal phonation. Besides, these models fail to establish mapping relationships between the vocal folds and the models. Addressing these issues, this paper proposes an inversion system for vocal folds models that can more meticulously characterize the nonlinear phenomena in vocal folds vibrations. To reproduce the nonlinear behaviors in vocal folds vibrations, this paper designs left–right asymmetry coefficients for the upper and lower masses specifically. Additionally, we introduce geometric Brownian diffusion as a lognormal statistical prior to reproduce the acoustic phenomenon of random jitter when vocal folds are diseased with a miniaturized parameter set. This paper adopts the Hippopotamus Optimization to establish an information interaction mechanism between the vocal folds and the model. The proposed method is validated using three databases. The MSE of the proposed inversion system on these three databases is 1.3%, 0.6%, and 1.0%, respectively, representing the best performance among the comparison methods. Simulation results demonstrate that the inversion system constructed in this paper, by designing a global nonlinear vocal folds model to characterize the vocal folds, achieves the simulation of nonlinear behavior in vocal folds vibration, establishing a high‐dimensional mapping relationship between the vocal folds and the model.