I had phoneme-based text-to-speech on a PC, later ported to a PS/2 for an IBM project.
Used the timer/speaker to generate waveforms.
Without the intent to begin with, it was a very good imitation of Arnold Schwarzenegger.
I added it to the database based management system I was writing for them, just to do text-to-speech of error messages.
Picture this (which actually happened) -
Leased Asset Manager -
alone in 20+ story building (plus a few security guards rattling around) -
Reviewing most recent pre-release version, alone, at night -
building shut down, not even a fan to be heard -
Hits one of the operator error messages:
Arnold speaks up with: "You are not allowed to do that, dummy."
After he changed his shorts, I got a middle of the night phone call to remove that feature.
(I am probably lucky they didn't remove me.
)
I might even have that code somewhere (it would be on paper if it survived this long).
It could probably drive an LED same as it drove a speaker.
And it also did not take a zillion bytes of ram (or much processor muscle).