Stylistically variable state space models for dynamic systems

摘要

A method models a dynamic system operating in a plurality of styles. Training data are acquired of the system while the system is operating in a plurality of styles. A generic model of the system is determined based on all of the training data. A variation-specific model is determined for the training data of each of the plurality of styles of operating. Stylistic degrees of freedom, style coordinates, and a specialization function are determined from the generic model and the plurality of variation-specific models. The style coordinates, the styles degrees of freedom, and the specialization function are applied to the generic model to produce specialized models. Sampling from the specialized model generates data simulating the system operating in an arbitrary style.