I did a whole bunch of articles about how web speech works, and the underlying speech subsystem - it's tricky because most text doesn't have any real semantic 'hints' and speech subsystems vary, so you can get "4x4=16" and "I drove a 4x4" sounding like 4-X-4 or "4 times 4" or "4 by 4" - kind of like with different human readers, maybe.
