You got a good one there.
I posted a corrected clip:http://www.kozco.com/audacity/gm_corrected.wav
That's two passes through the Cool Edit noise correction filter (not Audacity). I think the result is terrible and I know why. You picked a noise that has a lot of the same characteristics as the human voice. It's really rough to separate them.
I want to try this once again with a little change. You gave me an MP3 clip of the voice. MPEG1, Layer 3 (its real name) is a destructive compressor. It damages everything it touches. So the clip you gave me is 10:1 compressed and damaged compared to the one you have. I want the real, undamaged thing. I'm expecting a 3M to 7M clip, not a 700K clip.
You can save some data space by converting to mono and deleting the right channel which is silent in the clip I have. In audacity it's "Split Stereo Track" from the pull-down menu on the left. Send "Export as WAV...".