We have two opposing views (that people don't necessarily recognise consciously).
- We must modify the source(s) so that what reaches the listener's ear is as close to a specified (smoothed) frequency response as possible (but not time/phase response because this is considered just 'de-correlated chaos'). OR
- The speaker must be small (in relative terms) and neutral (anechoic frequency response, dispersion)
For (1), many, many different approaches can meet the goal including electronic EQ, phased arrays, reverse phase sources (even if that's just the backwave from the driver), resonators, acoustic panels. The aim is to patch together enough bits of frequency response material to fill the FFT bins to the right levels when making the measurement. A bit of a dip here? Change the value of that resistor, or aim that driver in a different direction. Ah, that
looks better.
Whereas (2) is so simple that it gives the impression that it's either going to be right (perfect in fact) or vastly inferior to the patch-and-mend approach. If a convincing example of (2) can be found that people seem to love the sound of, it's an indication that it's probably right.