Google Soundstorm had the best demo so far. It takes few seconds of original audio and continues it with the same voices. Just hearing those examples you wont figure out where original finished and generated one started.

Yeah, neural codecs are pretty amazing. The most incredible part is that they can do compression well across the temporal domain, something which has been non-trivial.

I'm also curious. A review of what's state-of-the-art today would be a great idea for a blog post. Just don't post it on medium.com please

Still Azure Speech in my experience.

ElevenLabs is better quality wise, but it's vastly more expensive. Azure Speech hits a really good price:quality ratio.