A computational method for updating a probabilistic model of an uncertain parameter in a voice production model

The aim of this paper is to use Bayesian statistics to update a probability density function (p.d.f.) related to the tension parameter of the vocal folds, which is one of the main parameters responsible for the changing of the fundamental frequency of a voice signal, generated by a mechanical/mathematical model for producing voiced sounds. Three parameters are considered uncertain in the model used: the tension parameter, the neutral glottal area and the subglottal pressure. Random variables are associated to the uncertain parameters and their corresponding p.d.f.’s are constructed using the Maximum Entropy Principle. The Monte Carlo method is used to generate the voice signals, which are the outputs of the model. For each voice signal, the corresponding fundamental frequency is calculated and a p.d.f. for this random variable is constructed. Experimental values of the fundamental frequency are then used to update the p.d.f. of the fundamental frequency and, consequently, of the tension parameter, through the Bayes’ method.