Conversational speech synthesis model for virtualized-egos