The whistles should be easy to isolate using spectral view- they will be at a higher frequency than the bulk of the music. It's a bit tedious to highlight each whistle as there will probably be multiple harmonics, but it's possible to reduce the volume in these areas with little effect on the part that you want to keep. You could also use spectral view to identify the frequencies of the whistles, then apply compression limited to those specific frequency ranges.
It's hard to say what would work best without an audio sample, but Audition 3 does have some good tools for this purpose.