In this paper we describe a (semi-automatic method for the characterization of the dynamics of the (flow) voice source in terms of variations in a set of descriptive parameters. The method uses the electroglottogram and oral flow as input signals. We use Hidden Markov Modelling (HMM) to search for typical patterns in the parameter set under different linguistic conditions. In this paper we focus on a discussion of the usefulness of HMM for modelling the dynamics of the source characteristics during the production of vowels and voiceless and voiced obstruents.