A comparison of local versus global imagedecompositions for visual speechreadingMichael