Faked Speech Detection with Zero Knowledge