Bear in mind that if this is ultimately for listening to with an aim of making it sound better, there's nothing worse for uncovering undesirable artefacts in lossy audio than messing around with it spectrally, especially if the filters involved are high-Q, ie, very narrow as that is almost guaranteed to make it sound far worse.
MP3 and AAC encoders walk a very fine line at low-to-mid bitrates when it comes to deciding which sub-band to assign more or less bits to in order to maintain apparent transparency, and they'll sometimes starve one sub-band of bits if others can benefit more in terms of overall psychoacoustic performance from using those same bits.
Bending the spectral response on playback past a certain point has a nasty habit of exposing this as audible unpleasantness.
In short, bend it too far and you'll break it.
