On modelling the dynamics of speech patterns