On the influence of text content on pass-phrase strength for short-duration text-dependent automatic speaker authentication