Towards modeling of speaker characteristics