I have tried audio CAPTCHA and could not succeed - and I have pretty good hearing.
I think that CAPTCHA has reached a practical end of life. It isn't just an accessibility issue, but a basic usability issue. There are essentially three ways to make text hard to read for a computer
1. add noise (like speckles and background patterns)
2. add distortion (bending words etc)
3. remove separation (push letters against each other or add bogus connections)
As I understand it, computers are already better than humans at solving #1 and #2 and rapidly closing in on #3.
In other words, we are just a few years away from the point where computers will be better than humans at solving CAPTCHAs. So it ends up being a reverse-Turing test. If you can solve the CAPTCHA, it's proof that you *are* a computer.
I expect that computers are not so good at voice recognition and these highly distorted voices are not as necessary as the highly distorted text, but I've read less about that.
In any case, we have definitely reached the point where the classic CAPTCHA is a poor way to protect a site.