DrDoc - 5:25 pm on Mar 2, 2006 (gmt 0)

I have implemented several CAPTCHA with audio files (never without audio). What I typically do is record a wav sound of all possible characters. The raw text string is stored securely on the server and retrieved by some form of session cookie or similar information. Then there's a link to a script which merges the applicable sound files before sending them back to the user. Has worked quite well.

Along with TJ's suggestion of merging that with additional information, I think it can be both accessible, usable, and nifty.

As far as the graphical representation goes... For those with no mobile impairments, I also offer the option of clicking on the image to generate a new graphic (same code, though).