I know there has been a lot of high level discussion here, which is fantastic, but also probably over most people's heads. Let me see if I can sum up for the layperson.
Ideally, a speaker will not add any colorations of its own to the sound being fed into it. The flatter the line shown in the graphs above, the less the speaker puts its own "stamp" (or coloration) on the sound coming out
of it. If you look at the M2, Salon2, F208, and F206 graphs above, you can see that the line is very flat. A flat line means that the speaker is not emphasizing any part of the sound spectrum. Moving from left to right on the graph, you start with bass frequencies and move up to the very highest treble. If you have a speaker that has such a flat line - or close to it - you can rest at least partially assured that what goes into the speaker is what is coming out (there's more to it to that, but it's a good starting point).
Now, let's look at speaker X, here:
You can see that there are massive peaks and valleys in this speakers frequency response. This means that certain frequencies are being emphasized or cut at the expense of others (the vertical axis of the graph shows relative volume of each of the frequencies, measured in db).
Now, think of an old school graphic equalizer:
Imagine putting a graphic equalizer like what's shown above in between your amplifier and your speaker. Ideally you would want to stay out of the way, and just pass what comes through it unaltered. This is what such a graphic EQ "curve" would look like - a simple flat line (using the free graphic EQ included in Audacity):
However, if you look at the response graph shown above, you can see that it has a frequency response that looks more like THIS graphic EQ curve (I did this down and dirty, for the purposes of illustration, but you can see that it closely reproduces what you see in the speaker measurement graph):
So, what you have now is a speaker that essentially acts like a graphic equalizer that's been permanently
placed into the signal path, applying that same EQ curve on everything that comes through it. And that's whether you like it or not - it's always there.
From the speaker response graph and the EQ above, we can easily predict that the speaker in question will have a big "hole" right in the 1500 hz to 3000 hz region, which is right in the upper range of the human voice:
The result would be a speaker that probably sounds "recessed" or lacking liveness and air in the upper vocal ranges.
Now, here is a speaker with a peak right around 2500 - 3500 hz:
Chances are this speaker would pretty dramatically emphasize that region instead, and we'd have a speaker with a particularly bright sound that emphasizes sibilants (s and t sounds) and would probably be considered "strident."
So, the obvious question becomes, why not just apply an opposite EQ curve to the speaker show at top, just like this:
One could be forgiven for thinking that such an EQ curve would "solve" all of that particular speaker's problems. After all, the above curve should "cancel out" all of the deficiencies (this is what some people think they can do with something like Audyssey). However, unfortunately, it's not so simple as that. As Harman engineer Tim Gladwin put it:
The reality is that the lumps and bumps in a loudspeaker curve come from many sources. Some can and should be fixed. Some should not. Often the fix simply does more harm. Anyone with experience in sound reinforcement or recording has played with an equalizer, and most notice that when you push a filter up too far or down too far, it howls. Once you hear the howl, then you start to hear it at lower levels, until you realize that it is there all the time. A filter is just another oscillator and it has its own sound. It always brings something to the table and usually what it brings is undesirable, so the less filtering the better. Generally we hear boosts more than cuts, so it is better to reduce a peak than to fill a hole. But making a speaker ruler flat, generally not a good idea. You could be coloring the sound worse than if you did nothing.
Specifically, many of the SPL features come from resonances in the cone, surround, dustcap, spider, basket or enclosure. These may or may not have uniform polar response. If they have uniform polar response, and are linear in level response, then it may be useful to attenuate a spike. However, if the SPL feature is not uniform in all directions, such as features caused by diffraction, then “fixing” it may make it worse in other directions.
So, the answer is to create a speaker that performs and measures flat to begin with. And that's what we have with both the M2 and the Salon2. The M2 has the aid of specific, targeted EQ that takes into account all of what Tim stated above to "tweak" the response curve, while the Salon2 does not - it's just an incredibly well engineered speaker to begin with
The above is incredibly
rudimentary and over-simplified, but hopefully lays some groundwork to form a basis of understanding. The "Spinorama" measurements from Harman tell us much more about how a speaker performs than what I laid out above, as they also take into account such things as off axis response, dispersion, and the directivity index. All those things together - not just the on axis frequency response I essentially discussed above - allow Harman to get to their 86% success rate in predicting which speaker will win the double blind listening tests. From the 5 Spinorama curves, they can make a good determination of what to fix and what to leave alone.