Yeah, looking at their web site, it uses an interior microphone, subtracts the music from the mic input, and then does both volume compensation and dynamic range compression based on what's left of the noise sample. Perhaps it even does some EQ fiddling, too, but I'm not sure if it tries to get a frequency sample of the noise level. Maybe it's just a bass loundness control.

Cool stuff, and theoretically do-able in the empeg all in software...