On the influence of rate, pitch, and spectrum on automatic speaker recognition performance