Comment by capitainenemo

It definitely uses whatever is available, but on review of the models (and there are quite a few) the ones I checked included instructions for integrating into linux speech engines.

What your distro uses by default might be espeak-ng, but you could use mozilla TTS or festival or piper or anything else. It makes sense to not bundle an entire system that repeats things that are available on all desktop environments and mobile these days, but instead call out to those from the browser functionality - especially given the enormous size of high quality speech models - not to mention the localisation issue. I think the TTS project was just to increase availability of high quality FOSS alternatives. Getting those into your distro is more of a distro packager thing.