An overview of text-independent speaker recognition: From features to supervectors