So this program by Harman was posted in another forum here; it's called How to Listen. It offers examples of different types of distortion that users can classify. Legit or no legit? It seems like it could be a very helpful tool.
If you look at research papers out of that group, there certainly is a case for it working.
So-called "trained listeners" (using this program) that they use for their research have been shown to outperform other groups. Look at a comparison of relative F-test values and some discussion here, which was prior to the release of the software:
Some of the later papers and blog entries have some more info on the performance of the trained listeners.