It actually understands roman numerals, to a certain extent. E.g. `say LVIII` will say "58". However, `say MCMLXXIX` speaks some gibberish that ends in the word "six", for some reason.
It also knows how to say numbers up into the trillions but not more than that (although I feel like it used to).
Other cute `say` tricks: muck around with OSX Default Speech Voice (to a siri-ish voice) and you can invoke that from the CLI.
I hacked together a little script for demo recording like:
START_FROM="$1"
STARTED=0
function transcript() {
ID="$1"
TO_SAY="$2"
if [[ ...STARTED || START_FROM && ID... ]]; then
STARTED=1
say "$TO_SAY"
fi
}
transcript "STEP001" "This is a test"
transcript "STEP001b" "of the emergency broadcast system"
transcript "STEP002" "This is only a test, if this was ..."
transcript "STEP003" "...etc..."
...and then I have a hardcoded `--output` which will then change the invocation to `say -o "$ID.wav" "$TO_SAY"` and output audio files.
That way I can iterate on voiceover scripts, eg: `./demo-script "STEP002"` => `./demo-script --output`
It's really helpful for iterating on pronunciations, eg: `transcript "STEP005" "In case of Four Hundred and Four errors..."`, I can just "skip to" and iterate against that line (and subsequent ones), or "skip back" and hear it in more of a flowing context.
...even if I don't end up using the `say`-generated audio, having a transcript (and even pacing) that I can just read through with my own voice is super helpful.
You can also change voices with -v. My favorite is "cellos" since it sings to you.