The Virtual World of Cerberus
While exploring autonomous evolution concepts for virtual worlds, we will present a new design Cerberus an avatar singer who can accompany a singer, perform alone and make his song evolves using simple past events. This 3D interactive facial animated avatar was made thanks to virtools software. The main originality of Cerberus is to develop his own melody by using learning machines and constantly improve his musical style and emotions. Cerberus is implemented using competitive learning rules to trained artificial neural networks in order to perform these self-improvements. Self-improvements is a key of our learning capacity, The challenge of building a virtual singer that could promotes his own improvisation is an open research field (Minsky, 2000).We will expose the difficulties of the synchronization in real-time between the voice and animation to generate the right emotion, also the difficulties of establishing a classification which could be in contradiction with the ontology of the musical fact. Also we will expose the necessity of developing avatars that use the amazing potentialities of spike-timing-dependent plasticity concept (Abraham, & Bear, 1996), that hence metaplasticity. This powerful concept enhance the potentiality of avatar design and give the impression that the avatar has a memory and simulate “feelings” linked to a context.