What about a lossy audio encoding system that uses information about the hearing profile of the listener to tune the encoding? For instance if you can't hear frequencies above a certain range then there's no need to encode the data for those frequencies. The software performs a hearing test to check which frequencies you can hear and create the profile. Obviously the files wouldn't be as good for swapping between listeners as generically encoded ones but I personally don't do that anyway. A further refinement would be to also tune for the playback environment. I'd like some files for playback in the car and some for through headphones at work. The first environment is much noisier than the first and the speakers are different. Perhaps a different encoding could exploit this to provide increased compression and fidelity for the portions that are actually going to be audible. Of course I'm not going to actually do anything with this idea :)