The speech quality is very good. It's not clear from the page that are they just adding styling to already generated audio or the audio is all generated using their own model?
They're generating the audio. They use a series of techniques to automatically generate metadata for speech samples in LibriSpeech for things like accent, recording quality, pitch, speed, gender, then use an LLM to format these tags into comprehensive natural language descriptions, leading to a more tunable model at inference time. This metadata generation pipeline is the key insight and what was missing from speech datasets unlike e.g. image datasets, which have obviously seen more rapid success.
Sounds like this is on a verge to revolutionise video game characters and YouTube dubs.