The Study Of Dynamics Of Voice Production

Many complex nature phenomena are governed by simple nonlinear laws. These nature phenomena are highly complex, and sometimes we can find stochastic property from these phenomena, so the intrinsic mechanism governing these phenomena can not be directly observed in experiments.Among these complex nature phenomena, the process of voice production is highly nonlinear. Rich nonlinear phenomena have been found in human speech, e.g., subharmonics, bifurcation, chaos and so on. The study of vocalization dynamics can not only be used to speech coding and synthesis, but also can help reveal the intrinsic mechanism of the process of voice production. We study the voice signal from two aspects, the one is global modeling the voice signal to analyze the dynamics characteristics of the vocalization process and the other is using the vocal model and neural model to study the complex process of voice production.In this paper, the term-ranking method is used to optimize the global model composed of radial basis functions. Using simulation and experimental data, the effectiveness of this method is well examined. After optimization, the prediction time of the global model is improved and the Bayesian information criterion of the global model is decreased. Then this method is applied to real voice signal, the result shows that the optimized global model can capture more predictable component (periodic pitch) in chaos-like voice data and simultaneously reduce the predictable component in the residual signal. Therefore, the speech coding and synthesis could be potential applications of this method.Based on the vocal-fold model and neural model, we build a global model. In our experiment, we use the voice signal, vocal-fold model and neural model to get the excitatory instructions. The research shows that simple neural excitatory instructions may generate complex voice signal.In this study, we analyze the voice process from two aspects to reveal the complex mechanism of the voice process.