I'm not up much on JAWS (or any other screen reader for that matter), but the ones that I've seen can either be set to read each screen automatically or they require that the user decide what to read by clicking a button, typing a key stroke, etc. to make the screen read to them.
The problem with sending keys is that (1) the software vendor could decide to change the required keys or (2) the user may be allowed to pick their own keys to press.
A much better approach would be to find out if JAWS (or whatever screen reader you use) has support for any type of API interface, scripting support, or maybe a command line interface that you could use instead. Another option might be to record the audio for each screen/function manually in some type of WAV or MP3 file and them play the files back at the right moments during the the install.
Sorry if this isn't much help. (If I happen to find anything more, I'll post it here.)