Speaker set identification through speaker group modeling