The problem with Audyssey Dynamic EQ

[-Matt-]

I have uploaded a video recently with calibrated mic/REW measurements of all Audyssey settings...

Thanks for sharing, the plots showing the effect of low frequency containment are interesting, and whilst I agree with you that the boost is too much, there are some other points in your video that I don't think are quite right...

1) I'm not sure it makes sense to compare with the Harman target curve; this is not the target applied by Audyssey. (Or did you already modify the Audyssey target to look like Harman - if so this could be another reason for excessive bass).

2) Your conclusion that bass boost starts from 1kHz and below comes from comparison with the Harman target; which isn't a valid comparison in this case. If you look at the plots in the OP it doesn't seem to do much at 1kHz, maybe it starts nearer 500Hz.

3) You can't really understand how Dynamic Eq and Dynamic Volume work from frequency response curves measured at only one volume level. The whole point of these features is that they apply different correction depending upon the volume level. This means that you have to be very specific about mentioning the volume used for your measurements. The point of the OP was to highlight that for the surround channels boost is applied at all frequencies (not just bass and top) this can be demonstrated with measurements at a single volume.

 

[OCA]

1) You're right, I 've edited Audyssey curve to replicate Harman but that only effects Reference setting, DEQ has the same exact boost with Flat setting, too.
2) Probably it's some roll off effect of the filter being applied but the slope really starts from 1kHz, the effect is around 2 dB @500 Hz and 4 dB @300hz, these are midrange frequencies!
3) DEQ & DV effect gradually decreases by volume and they have no effect at reference level. Measurements were taken -25dB from reference level as this is where I usually listen to the system.

The resolution of their filters is not high enough for correcting large peaks in the sub bass area so they dim the sub volume down and then try to compensate this with such gizmos. It's a fixation tool for their flawed correction system and I am sure it helps a lot of people, too. But this is not how it's being marketed and at least from an "audio science review" point of view, this needs to be identified IMHO

 

[Chromatischism]

Dynamic EQ (0dB) boost starts from 1kHz and goes all the way down gradually increasing to 12.5dB boost over the Harman curve in sub bass area. This is way too loud by any measure and if it's improving your SQ, there must be a huge dip in your system's mid & bass response. There's also a boost in the air frequencies which compensates for the over dimmed Reference curve HF response but the improvement does not justify the added excessive bass and ringing caused. I have uploaded a video recently with calibrated mic/REW measurements of all Audyssey settings if you want to see the details:

I'm not sure what volume level you are using, but unfortunately without that context, any measurements of Dynamic EQ or an equal loudness function are meaningless.

Anecdotally, DEQ has never sounded like "too much" in my systems, and I'm on speaker pair number 5, subwoofer pair number 4, and room number 5.

 

[Chromatischism]

The resolution of their filters is not high enough for correcting large peaks in the sub bass area so they dim the sub volume down

Wrong...their bass corrections are high resolution. See: https://www.avsforum.com/threads/wi...r-than-sonos-arc.3250986/page-4#post-61939848

 

[OCA]

I'm not sure what volume level you are using, but unfortunately without that context, any measurements of Dynamic EQ or an equal loudness function are meaningless.

Anecdotally, DEQ has never sounded like "too much" in my systems, and I'm on speaker pair number 5, subwoofer pair number 4, and room number

Assuming reference level is 105 dB's peak in any given channel full range, the measurements were taken at 80dB.

If you leave everything to Audyssey automation then these gimmicks improve the overall sound but there will still be minor (and to me major) flaws. Channel volume levelling is done only at 1 kHz, subwoofer timing is always off, midrange is heavily boosted and I won't even start on midrange compensation. If you want precision and descent FQ response, it's still possible with Audyssey with some extra work but then you should defo turn all Audyssey tools off.

 

[-Matt-]

The resolution of their filters is not high enough for correcting large peaks in the sub bass area so they dim the sub volume down and then try to compensate this with such gizmos. It's a fixation tool for their flawed correction system and I am sure it helps a lot of people, too. But this is not how it's being marketed and at least from an "audio science review" point of view, this needs to be identified IMHO

It seems that you may have misconceptions about the aim of these features...

Audyssey offers full range frequency response correction and later versions have quite good resolution with their filters (even at bass frequencies). This was an area where previous versions of Audyssey fell short, so you may still find criticism of this online. (The reason that you still see bumps in your frequency response is because not all room mode effect can be fixed by Eq, and certainly not at all positions).

Dynamic Eq is separate to the target-matching, frequency response correction part of Audyssey. Its aim is to modify the target curve depending on the master volume. Aiming to keep perceived loudness constant (by boosting bass when master volume is low).

 

[OCA]

Wrong...their bass corrections are high resolution. See: https://www.avsforum.com/threads/wi...r-than-sonos-arc.3250986/page-4#post-61939848

16,000 (16384 actually) is the FFT length, not the number of taps. They use 1024 taps for subs (and only in the top version) and 512 taps for the rest of the system.

And FYI even 16000 taps is far from being adequate for sub bass correction. Look how it totally misses the 32Hz peak:

 

[-Matt-]

@Serkan can you let us know the source for this information?

What is the plot above showing? What software is it taken from? If this is intending to show Audyssey filters, how were these extracted from the AVR?

 

[OCA]

@Serkan can you let us know the source for this information?

What is the plot above showing? What software is it taken from? If this is intending to show Audyssey filters, how were these extracted from the AVR?

This is a just a sample FIR filter I created in rePhase to illustrate inadequacy of 16000 taps, would vary a little depending on which windowing is used though.

 

[-Matt-]

Your estimate of resolution based on rephase may not accurately reflect what Audyssey does; they claim to do something "unique" with their filters.

https://support.denon.com/app/answers/detail/a_id/380/~/audyssey-multeq-xt---low-frequencies

 

[OCA]

It seems that you may have misconceptions about the aim of these features...

Audyssey offers full range frequency response correction and later versions have quite good resolution with their filters (even at bass frequencies). This was an area where previous versions of Audyssey fell short, so you may still find criticism of this online. (The reason that you still see bumps in your frequency response is because not all room mode effect can be fixed by Eq, and certainly not at all positions).

Dynamic Eq is separate to the target-matching, frequency response correction part of Audyssey. Its aim is to modify the target curve depending on the master volume. Aiming to keep perceived loudness constant (by boosting bass when master volume is low).

In an ideal World, this would be the intent and it would be great but I really think 12 dB is way too much. And after your comments I started thinking it might be the effect of the Minidsp 2x4 I am using for subs. I will check for that in my free time.

But Audyssey came out in 2004, was a game changer at the time and all and I have been a very loyal user ever since (never switched to Dirac) but even Harman curve was created in 2012. The technology evolved very fast in digital signal processing and if noone criticizes them, they will still be launching editor apps on the same hardware by 2030.

Noone in the World has the room measurement data they have, imagine 8 measurements of 12 speakers by every user multiple times and since decades. Imagine what AI can achieve with that many samples. I believe they can do a lot better if they are pushed a little bit

 

[OCA]

Your estimate of resolution based on rephase may not accurately reflect what Audyssey does; they claim to do something "unique" with their filters.

View attachment 228281 https://support.denon.com/app/answers/detail/a_id/380/~/audyssey-multeq-xt---low-frequencies

They may be using rectangular windowing with some tricky centring but still there's really a limit to what you can do with 1024 taps. In their defence, more taps would cause delays and although it's a sound correction system, it needs to cope up with video frames at all times.

 

[-Matt-]

At the risk of leading the thread further away from the topic of Dynamic Eq...

I'm still curious where the figure of 1024 taps for subwoofer correction is from? Is it verified as correct? What does this translate into as a frequency resolution at 20Hz?

I think Audyssey prefer not to release the actual number of taps (probably because this could lead to unfair comparisons given their non-standard implementation). However, from this table:

...from this thread, it is claimed that MultEQ XT32 has 512x the resolution for the subwoofer (relative to the most basic version of Audyssey I believe). Does this mean that the basic version only has 2 taps?

This graph from a little further down the same thread shows the Eq corrections (meaured directly at the pre-out) for different Audyssey versions:

(This is not a speaker response, it is the Eq modified output, and so is not supposed to be flat). You can see that both of the XT32 curves are able to produce a fairly narrow peak at about 25Hz. (...and perhaps could have produced a sharper peak if necessary to achieve the target curve).

 

[OCA]

At the risk of leading the thread further away from the topic of Dynamic Eq...

I'm still curious where the figure of 1024 taps for subwoofer correction is from? Is it verified as correct? What does this translate into as a frequency resolution at 20Hz?

I think Audyssey prefer not to release the actual number of taps (probably because this could lead to unfair comparisons given their non-standard implementation). However, from this table:

...from this thread, it is claimed that MultEQ XT32 has 512x the resolution for the subwoofer (relative to the first version of Audyssey I believe). Does this mean that the first version only had 2 taps?

1024 taps has exactly 16,384 FFT length, that's what I am certain of and I think only "MultEQ XT32+SUB EQ HT" has 1024 taps for the sub. 2 taps is possible, will work like a 2 band graphic equalizer (not parametric) so those multipliers in the table are most possibly number of taps. But as you said, they don't disclose.

I know a lot about their measurement impulses (16,384 data points, from 0 to 24000 Hz hence 48kHz sampling rate, 8 bits precision, 0.3 seconds long, multiple repetitions to eliminate HF reflections rather than using frequency dependent windowing) because I could extract them from the app but no clue on their convolution impulses, these are hidden in heavily encrypted very large files

 

[-Matt-]

I know a lot about their measurement impulses (16,384 data points, from 0 to 24000 Hz...

I understand that they have the same number of taps for subwoofer correction as for the full range channels. I thought these would cover a narrower frequency band (to give higher resolution), but perhaps that is just wishful thinking?

Anyway, lets agree that Audyssey has plenty of some room for improvement! I hope they are listening and are currently working hard to bring some meaningful improvements to their offering. If Audyssey ceases to be a positive selling feature (arguably Dirac is already more desireable) then they may find that they are dropped by the AVR manufacturers, which is potentially fatal to the company.

 

[OCA]

At the risk of leading the thread further away from the topic of Dynamic Eq...

I'm still curious where the figure of 1024 taps for subwoofer correction is from? Is it verified as correct? What does this translate into as a frequency resolution at 20Hz?

I think Audyssey prefer not to release the actual number of taps (probably because this could lead to unfair comparisons given their non-standard implementation). However, from this table:

...from this thread, it is claimed that MultEQ XT32 has 512x the resolution for the subwoofer (relative to the most basic version of Audyssey I believe). Does this mean that the basic version only has 2 taps?

This graph from a little further down the same thread shows the Eq corrections (meaured directly at the pre-out) for different Audyssey versions:

(This is not a speaker response, it is the Eq modified output, and so is not supposed to be flat). You can see that both XT and XT32 are able to produce a fairly narrow peak at about 25Hz. (...and perhaps could have produced a sharper peak if necessary to achieve the target curve).

EQ modified output only means frequency response after Audyssey correction, you don't need a pre-out for that, I measure these every day with REW and Umik-2. XT32 flat and XT32 ref graphs have been applied VAR smoothing. The blue one has variable frequency dependent windowing and you can't do that in REW unless you use merge, this is probably done in Acourate/Audiolense. Anyway, this has nothing to do with MultEQ version resolutions, it's three measurements smoothed differently to give that impression.

They are evidently Audyssey calibrated responses though. You can tell from the odd HF dip and the state of the art midrange compensation in Reference compared to Flat

 

[-Matt-]

EQ modified output only means frequency response after Audyssey correction, you don't need a pre-out for that, I measure these every day with REW and Umik-2. XT32 flat and XT32 ref graphs have been applied VAR smoothing. The blue one has variable frequency dependent windowing and you can't do that in REW unless you use merge, this is probably done in Acourate/Audiolense. Anyway, this has nothing to do with MultEQ version resolutions, it's three measurements smoothed differently to give that impression.

My understanding was that these are not microphone measurements, electrical masurements were instead made direct from the pre-out. This removes errors that would otherwise be introduced by the speakers, room and mic.

With no correction the frequency sweep above would be almost perfectly flat, with correction it follows whatever profile Audyssey decided was needed for this room. The point is that those corrections for XT32 can contain fairly narrow peaks even at low frequencies, demonstrating its resolution, whilst XT had very poor resolution in the bass. (Of course XT32 may actually be able to create even higher resolution features, these may just not have been warranted by the room).

I think they were measured using REW and have no smoothing applied (no icons in the legend, although that could be due to an older version of REW).

 

[OCA]

I understand that they have the same number of taps for subwoofer correction as for the full range channels. I thought these would cover a narrower frequency band (to give higher resolution), but perhaps that is just wishful thinking?

Anyway, lets agree that Audyssey has plenty of room for improvement! I hope they are listening and are currently working hard to bring some meaningful improvements to their offering. If Audyssey ceases to be a positive selling feature (arguably Dirac is already more desireable) then they may find that they are dropped by the AVR manufacturers, which is potentially fatal to the company.

What's happening in Audyssey is quite typical. When a tech company grows large enough, marketing people take over management from engineers. From there on, surveys determine the strategy and average Joe rules the surveys, not the enthusiasts. If the questionnaire above this topic is any measure, we'll soon get more sophisticated Dynamic EQ

 

[-Matt-]

I'd settle for it just not being broken!

If Dynamic Eq did the same on the surrounds as it does on the fronts I'd be happy.

 

[Chromatischism]

16,000 (16384 actually) is the FFT length, not the number of taps. They use 1024 taps for subs (and only in the top version) and 512 taps for the rest of the system.

And FYI even 16000 taps is far from being adequate for sub bass correction. Look how it totally misses the 32Hz peak:

View attachment 228280

I'm not sure what this is showing, but it's not Audyssey.

Here is what Audyssey does for me.

Starting with some older measurements, speakers only, before and after:

Considering what it was given to work with, I'd say it did a darn good job.

Testing crossover slopes with the front speakers:

Flat amplitude through the bass region, save for the 95 Hz cancellation it can't do anything about (and intelligently does not try).

Here's one from years ago when I was testing different crossover slopes on the sub side. Bass was flat:

A little more recently, sub+speaker:

And most recently, center+subs, before and after:

It is easy to see how fine-grained the corrections are by looking at the black (after) vs blue lines with no smoothing. All those little hairs and spikes that were not there before are evidence of the resolution of the filters. Again, this is no smoothing and focused only on the 10-300 Hz range, so very revealing of problems.

These are all in-room measurements. There was no curve editing done in any of these measurements. I don't even believe editing the curve is necessary in most cases. Just press the button and go.

The only "tweak" I'm doing is adjusting the sub distances to phase match with the main speakers, but it only makes a small difference in my case. It's barely audible.

So, I have disproven your theory that Audyssey does not have a lot of resolution with measurements of my system over the years.