I tried to replicate their demo text but it doesn't sound as good for some reason.

If anyone else wants to try:

> Kitten TTS is an open-source series of tiny and expressive text-to-speech models for on-device applications. Our smallest model is less than 25 megabytes.