Text-to-speech voices sound pretty awkward. To make them sound less awkward, you have to either tweak the software for all sorts of special cases or pre-record a bunch of phrases. Both of these take a lot of manpower.

Fonts are the same, except they're used by many more people and last much longer than that brief moment when your text-to-speech software stumbles on an uncommon combination of words.

I'm sure that the technology will rapidly improve as more and more Chinese fonts are needed, but as long as AI remains inferior to humans in some way, I don't see it getting all automated.