It should be called video to speech or something, I would have be really surprised if such thing exist already.
I know there is this 'device' project (I forget the name already) that was trained to interact with web UI, so a virtual assistant could buy or reserve stuff for you on the net, so I guess you make an IA learn an interface.

But more seriously it wouldn't be that helpful for time lapse modelling video, it would only describe what's happening on the screen but not the reasoning behind.
I came across a commercial zbrush video/tutorial like that, where the 'instructor' was commenting his 4 hours sculpting ... it was very depressing to watch.

Those time lapse modelling session does have their merit, but only if you already have a good base knowledge.