It wouldn't necessarily need to be realtime - if you're using the device to build/tweak a "fingerprint" of background noise or required corrections, you could do that analysis with samples and periodically push the resulting filter profiles back to the earpiece (or something similar) after the phone has done the number crunching to generate them.