I was impressed by the demo and tried it. But I must say it’s not very straightforward to use. It can only handle about half a minute of audio, so the input needs to be split up and recombined, for which we didn’t have any examples as of yesterday. And it speaks very very fast and chops off things once the parameters are wrong. Sometimes it also did that for me with short input. And the voices weren’t a 100% consistent match, even with the audio snippet in front to guide it.

I’d say it definitely needs more polish before being usable for an application. It’s impressive, though.