even though this sort of "feature" is highly sought after by the general public, it is very difficult to solve. even if you did lots of manual tweaking using filters, eq bands, etc. the results would still be dubious at best. if you are ok with that though, two suggested techniques are:
- eq filtering to reduce the amplitude outside the (approx.) 200-4000Hz voice spectrum
- sampled noise filtering (sample the music as noise and then do a reduction based on that sample)