New method for mathematical modelling of human visual speech