System that sounds so real it is easy to mistaken it is not live

@asctim my Dad was a trumpet player and I've been conjecturing for years why no recording of a trumpet (and worse, string sections) doesn't sound anything like live. 

My theories are:

1) Dispersion pattern, as you noted. 

2) Mismatch between speaker type and mic type

3) Trumpet has a DC component to the outflowing air column that may create complex phase variances microphones can't capture or speakers can't play

It is a constant annoyance when guitars sound 8 feet wide, violin sections sound like a bad synthesizer, pianos sound like they are mounted vertically on the front wall, etc. and I agree recording technique is largely to blame. 

I've considered building a pair of dodecahedron mics and matching speakers to see if that works. But 24 channel stereo just isn't something current media and gear can handle.

@esputnix my one jaw drop moment was taking a tour of Mechanics Ha in Worcester, MA. First we were in the hall itself, then we went upstairs to their highly customized remote listening area with B&W 801s just outside the mixing booth upstairs. The walls were specially treated and the room was packed with people. Because we had just left the hall our ears were accustomed to its acoustics, so when they played a live recording it seriously sounded like w and a requsst in right in front of the orchestra. It was magical. I've never heard anything like it since, but if you have the opportunity to travel, ask them if they can arrange for a tour so you can hear what I heard. 

They played 

@lewm , thanks Lew. That is a common subwoofer mistake, crossing over too low. The makers of commercial subs using just a low pass filter suggest that but, it does not clean up the main speaker as much. Info below 40 Hz is relatively rare. If you can see the cones of the subwoofers moving then Doppler distortion is at play not to mention the non linearities you get into on far excursions. If you play test tones you can easily see the cones moving up to about 100 Hz. This is double important on ESLs because the driver is full range. When dealing with a dynamic speakers it will only be the range of the woofer that is cleaned up. This is still substantial in most instances. This mandates a complete two way crossover. I am also talking about digital crossovers which are a benefit as delays can be adjusted so that the signal arrives at the listening position at the same time in phase. The best location for dipoles in the room is never the best location for subwoofers. It is not easy to make subs blend in with 8 foot tall ESLs. It is not the dipole nature of ESL that creates the conflict but rather the line source characteristics. It requires a line source subwoofer especially if crossing up higher. As I previously mentioned Dr West is aware of all of this but he has to sell loudspeakers to survive and many people do not want to get involved in loudspeakers that require subs. Jim Strickland of Acoustat was the same way, maybe worse. He stubbornly refused to admit his speakers benefited from subwoofers. 

We once calculated that my speakers have about 400 in.² greater radiating area then do yours. Perhaps that’s where our results differ, because my speakers are not perceptibly distorting in anyway at 100 Hz. And there is the added value of coherence and continuity with higher frequency music. Now you are saying there is no point in reproducing music below 40 Hz, because there is not much information. Isn’t that contrary to your previous assertions about your system going down to 15 Hz or thereabouts? Why do you bother to be concerned about frequencies below 40 Hz, if they are irrelevant? I say this Not as someone who believes that frequencies below 40 Hz are irrelevant.

I remember a blind test in the 80s where blindfolded listeners compared live performances with recordings of the same music in the same space.  The results were strong; most listeners were able to distinguish easily and reliably.  Even $1,000,000 systems have a long way to go.

These tests should be repeated.

It is also easy to run blind tests of components.  Set up two identical complete systems in the same space.  Insert the two components to be tested in the systems.  Play the same recordings A B X, where blindfolded listeners are to identify X as A or B.  Repeat say 10 times.  Score the number of successes against the probability of the result being chance.   For sticklers re-run the test with the common system components swapped out.  Golden ears should be very afraid, but these tests should be run to establish whether they have any worthwhile credentials.