Text normalization for reading Wikipedia articles

[xenotropic]

So I ran a few Wikipedia articles through Tortoise, just to see how it does, and because it does seem like a real problem domain where tortoise might be used. You can listen here to the outputs, which is a podcast RSS feed. It does quite well, but the main gap seems to be a need for "text normalization". That is, making like "300m" into "300 meters" and "$1.2 billion" into "one point two billion dollars". Listening to Tortoise read "65 kn (120 km/h; 75 mph)" in the Hurricane article is quite disorienting!

It seems like NVidia's NeMo is the main open-source option I'm seeing around Github, although it is slow, has a lot of dependencies, and just throwing in some articles it came out with at least some errors ("kmh" into "Kilometers per H" in particular). Before I dive into learning about it, I want to ask: are there any other candidates to look at that anyone is aware of? It seems like there was some big Google Kaggle on text normalization a few years ago, but at least on a quick look they seem optimized to the contest itself and not adapted for ready application to real world use. Thanks -- Joe

[neonbjb]

Pretttty sure that chat gpt or gpt 3.5 can do text normalization for you. I'd look into that.