Effects of mouth-only and whole-face displays on audio-visual speech perception in noise: is the vision of a talker's full face truly the most efficient solution?