> It is a tool I am working out, trying to automate my teaching job. I seem to remember you mentioned using some mac os program "Say" to generate the speech. If it could be replaced with e.g. the text-to-speech as offered by google on their translate.google.com page, it 'd become cross-platform and multi-lingual... (apparently someone even made a python module for that: gtexttospeech - disclaimer: I haven't tried it out.)
I think declaratively generating screencasts is a seriously cool idea, a bit like the screencast equivalent of "slide generation via LaTeX".