• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Verification of Audyssey’s XT32 results

Deafboy

New Member
Joined
Oct 9, 2021
Messages
4
Likes
21
My first post on ASR!

Recently I purchased a Denon AVR-3700H receiver based on amirm’s review of the unit and also his positive impression of the Audyssey Room Equalization system (reviewed separately) that is included in the Denon AVR-3700H. One aspect that piqued my curiosity is his opinion that the Audyssey’s room corrected response shown on the app is too good to be true (“wishful thinking”). The “after” curve shown in the app is undoubtedly a simulation based on Audyssey’s series of measurement. Jeff Clark, Director of Software Engineering at Audyssey, mentioned in a Youtube video that if one combines the 8 measurements in the exact positions where the calibration was done with the Audyssey processor enabled one would obtain the “target curve” shown in the “after” section of the app. I wondered how close this simulated response was to the real response after the room had been “Audysseyfied”. I did an experiment to verify this and would like to share the results. I’m also new to ASR so if this has been done in the past, and I am reinventing the wheel then my apologies.

Here’s a description of my experiment. My loudspeakers are late 1980’s Canadian-made Energy 22.3. I first affixed a measurement microphone right next to Audyssey’s microphone that came with the Denon (see picture below). Then for each of the 8 measurements in Audyssey’s calibration procedure I noted the position of my tripod with marks on the floor. Once the whole Audyssey procedure was done I then re-measured in those 8 positions each speaker with my microphone and REW software (Room Equalization Wizard) with Audyssey disabled (Denon’s “Direct” mode) and then enabled. So for a pair of speakers this represents 32 measurements done with REW. I used REW “RMS Average” function to combine the 8 measurements corresponding to a set in the experiment.

The results are quite revealing. The picture below shows the screen grab of the Audyssey app showing the “before” and “after” results with the right speaker. I overlaid my results taken with REW in curves below Audyssey’s for comparison, being careful to precisely scale the frequency and amplitude axes. Using a smoothing factor of 1/24 octave to best match my results to Audyssey’s, we see that Audyssey results of the un-corrected are surprisingly close to my results (curves on the left). The “After” results I obtained are certainly more jagged than Audyssey’s simulated results but I consider these impressive nevertheless.

Comments welcome. :)

1633816788176.png


1633815280849.png
 
Last edited:
Nice, thx for taking the time to post!

Edit: can you post your rew before and after with the same smoothing?
 
I’ve always wondered about this. Awesome someone decided to come with facts and put in the work. Thanks for doing so. Did you measure more than one channel? I’d love to see the subwoofer result.
 
can you post your rew before and after with the same smoothing?
I see what you are asking for, but he needed to post the unfiltered REW "before" to compare with the Audyssey mic's "before".

Remember, the thread is not about what Audyssey does, but whether Audyssey's predicted "after" curve is realistic.

A thread on what Audyssey does is really a very different thread.

cheers
 
can you post your rew before and after with the same smoothing?

Yeah, with the same smoothing. Will give a better pic of what Audyssey cal is doing.
These are the curves I posted in my first post. Both of my curves uses the same smoothing (1/24 octave) for each measurement. Is this what you are looking for?
 
Last edited:
Did you measure more than one channel? I’d love to see the subwoofer result.
I decided to keep it simple for now, so just the main speakers. Perhaps a later experiment will have the subwoofer.
 
Seems like about 3-5x the peak/dip decibels in the measured curve compared to the predicted curve (~1-2db dip becomes ~4-5db), and the tiny jags which are too small to tell how many decibels they are become 1-2db in the measured response. So I'd still say the prediction is a fantasy and doesn't tell you much about what Audyssey did.

The only useful information I seem to be able to read from the "after" curve is where were the problems that are so extreme that Audyssey didn't even try to flatten completely.
 
They definitely are still using some kind of smoothing... which I find a bit dishonest because it's not very clearly disclosed.

As to the full-range correction: Well, I also have doubts about the need to completely fill-in the xo dip. That itself alone could potentially exaggerate resonances/distortion.
 
There is smoothing (they mention it themselves) but it doesn't seem to explain the differences, as you can see the resolution of the tiny jaggies on the "after" graph, so it's not smoothing that causes the discrepancy.
Fact is most "after" graphs I've seen for XT32 are almost completely flat, with very tiny jaggies, with the occasional peak/dip that it admits it couldn't fix (which in reality is much bigger than predicted).
By the way, for XT the "after" and the real result, at least for the lower frequencies and subwoofer, seem to be more in-line, but that's easy considering it hardly does anything to the low frequencies of the speakers, and does limited corrections for the subwoofer.
 
Back
Top Bottom