Did you try this with a single point calibration though? That is the only easy way to eliminate the effect on averaging methods. If Audyssey decided not to fix something due to variations between the different positions, it'll appear as if it missed something that could be corrected when in fact it was intentional. Unless you rule out the measurement locations and averaging methodology, you don't really have any proof to your claim, which I find quite odd - If Audyssey couldn't correct something the first time around, why would adding a peak/dip in the target curve make it correct it? At least according to how Audyssey is expected to work, getting that extra correction with modifying the target curve will only correct things that Audyssey could correct, but decided not to, and you set the target curve to bypass Audyssey's decisions. If I'm right, then it's probably just better to revise your mic positions when running Audyssey calibrations. If you have the app you can probably try various combinations of positions relatively easily, and see the effect.
I don't see where you posted results that were specifically titled as single point Audyssey calibration compared to a single-point REW sweep?
Also you'd need to make sure the difference isn't purely because of a difference in the microphones or their positions even with just a single point.
Tough crowd here. Alright I just did a single point calibration experiment, Audyssey on/off and with/without REW correction. Front speakers only, subwoofer disabled. Receiver is Denon X3600, speakers are Philharmonic BMR v2s.
To start with, this is what the Audyssey mic looks like at the MLP position in my setup (the mics are pointing up, the camera angle shows it tilted). The UMIK-1 is right next to it for comparison (but not present during calibration). When I'm taking REW measurements, the UMIK-1 is placed confidently within 1 inch of the Audyssey mic position, and the Audyssey mic is removed from the setup (tripod remains for the sake of keeping things as consistent as possible).
Ok, so on to the measurements. Here I am looking at overall smoothness of the response and how well the left/right channel match (which impacts imaging). I tried to quantify the average left/right SPL error with linear and log weighting (the log weighting will weight bass mismatch more).
First we have the Audyssey off measurement, showing typical room effects and significant bass frequency mismatch:
Next we have the Audyssey measurement, showing great improvement of bass modes and significant improvements to left/right channel matching. However, strangely there is a high frequency rise and some peaks/valleys in the mid/high freq response. In the MultEQ-X "Design Target Curve" tab, I have "Theater high frequency roll-off 1" enabled, so it's very strange I'm getting this rise. I've run probably a couple dozen calibrations with different configurations and speakers, and I've always seen this post measurement. I
suspect my Audyssey microphone is not well calibrated. I have ordered a calibrated Audyssey microphone so I can determine if that's the issue (will check this weekend). I do not believe my UMIK-1 is miscalibrated; I've tested it against a different UMIK-1 and they matched well (and I am using the 90 deg cal file).
Below is the Audyssey correction after putting in filters from REW. The target curve I used here was a flat response for a "bass limited speaker" with a 6 dB/oct roll-off at 30 Hz. This provides additional smoothening of the response, and some more improvements in left/right matching for higher frequencies.
So my conclusions are that Audyssey is (1) trying to achieve some strange non-flat target curve, which may be a microphone problem, and (2) there is some potential correction left on the table that Audyssey could correct, but doesn't.