that might be the next AI level of compression of just about everything. Very lossy, but very small.

It would be.. spooky. Just like the lady who voiced Siri just had to record sounds (letters and combinations of letters) that can be combined for text-to-speech, one could do that for podcasters...

But then it means you can make your podcaster say anything you want.

Combined with a good text compressor you could get > 100 hours of speech on it. http://www.mattmahoney.net/dc/text.html