Abstract: A system for generating an animated image of at least the head of a character which is speaking, with the face of the character having visible articulation or expression matching the words being spoken, is described. An input stream of marked up text comprising text to be spoken by the animated character and instructions representing behavioral features are combined in a modified text stream which includes randomization instructions to create random behavioural features. The modified text stream is processed to produce an audio stream representing the text being spoken and a sequence of phonemes corresponding to portions of the audio signal. The phoneme stream is mapped by phoneme to viseme mapping means to produce a stream of morph targets. The morph targets are, in turn, combined with further random movement instructions and these are then processed in an animation module to generate an animated image of at least the character's head.