Vocoding

In order to model speech, we need a parametric representation of it. This might be done using a source filter model, or something more general.