The guy has pretty big constraints on his setup: Voice Cloning AND open source …

At the top of the leaderboard is Kokoro. Kokoro is an amazing model! Especially for a modest 82 Million (!) parameters and a mere 360 MB (!). However, like many models in this leaderboard - I can’t use it - since it doesn’t support voice cloning.

They left out VibeVoice, the leading model for voice cloning that is intended exactly for the described use case?

Yeah, I don't want anything that needs internet or some sort of AI model.

I don't even mind older methods, so I guess DECtalk is probably the best option there is (because newer simple TTS is just as robotic but with worse aesthetic on the default voices) unless there are newer speech synthesis projects I don't know about.

Plaintalk is quite the aesthetic too, but I imagine IP law keeps it out of free software.

Ideally I'd like a synthesis system that has a mix of different types of voices. Old-style robotic, more sleek robotic, semi-human sounding etc. Sounding good is more important than realism though, a non-convincing voice that can still pronounce things is better (some of that may be down to phoneme input as well, but it would be nice to have normal text input 90% of the time).