Automatic parameterization of the voice source is dependent on the availability of a suitable parametric model of the glottal flow pulses. Models like the Liljencrants/Fant (LF) model may not be rich enough to accurately match all those characteristics found in the results of inverse filtering. This may result in noisy estimates of the model parameters if flow waveforms are approximated that have features not accounted for by the model. In this paper an overview is given of the different types of source waveform that were obtained through automatic inverse filtering. We point out a number of features that we think are relevant and that are difficult to describe with the LF model.