A method, performed by a user equipment device, for text-to-speech conversion entails sending to an articulatory model server exterior facial structural information of a person, receiving from the articulatory model server at least a portion of a predefined articulatory model that corresponds to the exterior facial structural information, the predefined articulatory model representing a voice of a modeled person who is different from the person, and generating, based at least partly on the predefined articulatory model, speech from text stored in a memory of the user equipment device. Furthermore, a method of configuring text-to-speech conversion for a user equipment device entails determining at least a portion of an articulatory model that corresponds to exterior facial structural information based on a comparison of the exterior facial structural information to exterior facial structural information stored in a database of articulatory models.