Meyer Sound Amie Monitor Review

[dasdoing]

well, mixing/mastering engineers use phantom centers

 

[OCA]

Score is 5.1 and would be 7.3 with a perfect subwoofer (which is likely the case with the Amie sub).

The speaker is very well optimised and I would not use an EQ with it.

With eq, it looks better on the on-axis and listening window but I am not convince by the change in the PIR and you can see the negative effect in the histogram. I would need to listen to it to see if that makes a positive and audible difference.
With EQ, score would increase marginally up to 5.6 (resp. 7.8 with sub).

View attachment 315908

EQ for Meyer Sound Amie computed from ASR data
Preference Score 5.15 with EQ 5.61
Generated from http://github.com/pierreaubert/spinorama/generate_peqs.py v0.25
Dated: 2023-10-01-14:55:41

Preamp: -2.7 dB

Filter  1: ON PK Fc    58 Hz Gain -1.52 dB Q 2.99
Filter  2: ON PK Fc    95 Hz Gain +0.99 dB Q 1.87
Filter  3: ON PK Fc   260 Hz Gain +1.80 dB Q 2.99
Filter  4: ON PK Fc   560 Hz Gain +1.23 dB Q 2.98
Filter  5: ON PK Fc  1429 Hz Gain +1.20 dB Q 2.95
Filter  6: ON PK Fc  4051 Hz Gain +1.01 dB Q 2.94
Filter  7: ON PK Fc 10341 Hz Gain +2.65 dB Q 0.53

I am a bit concerned about the meaningfulness of the "with EQ" scores of tested speakers. Who decides which parameters should be used to optimize the EQ for each speaker? Does a flatter frequency response mean better sound? What about the phase shifts and delays introduced by these IIR filters, and the changes to the peak energy times of these frequencies?

I understand that the same optimization parameters are used for all speakers to ensure comparability. However, it is possible to auto-EQ every frequency response to be perfectly flat (or on target per se) with the right parameters and method, and this would most probably just sound awful. But there will be cases where some very sharp filters will help improve the response of a certain speaker with phase anomalies in the design, while totally ruining it for another.

The seemingly harmless filters above with a maximum bandwidth of 3 (which I believe to be proportional Q) and a maximum gain of 2.65 dB will still cause more than 10 degrees of phase shifts at many frequency bands (dotted line below). Depending on the actual phase response of the particular speaker, these phase shifts may help or worsen the sound of the speaker. Again, this is a very mild EQ and a few ms of delays caused by 10 degrees are inaudible in practice but I see a lot more aggressive EQ parameters making their way to spinorama scores from time to time.

The author's EQ generating code is one of the best and most sophisticated I have seen and comes with no less than 50 optimization paramaters:

spinorama/generate_peqs.py at master · pierreaubert/spinorama

A library to display and compare spinorama (speakers measurements) graphs. - pierreaubert/spinorama

github.com

But that's also my point here!

 

[ernestcarl]

I am a bit concerned about the meaningfulness of the "with EQ" scores of tested speakers. Who decides which parameters should be used to optimize the EQ for each speaker? Does a flatter frequency response mean better sound? What about the phase shifts and delays introduced by these IIR filters, and the changes to the peak energy times of these frequencies?

I understand that the same optimization parameters are used for all speakers to ensure comparability. However, it is possible to auto-EQ every frequency response to be perfectly flat (or on target per se) with the right parameters and method, and this would most probably just sound awful. But there will be cases where some very sharp filters will help improve the response of a certain speaker with phase anomalies in the design, while totally ruining it for another.

The seemingly harmless filters above with a maximum bandwidth of 3 (which I believe to be proportional Q) and a maximum gain of 2.65 dB will still cause more than 10 degrees of phase shifts at many frequency bands (dotted line below). Depending on the actual phase response of the particular speaker, these phase shifts may help or worsen the sound of the speaker. Again, this is a very mild EQ and a few ms of delays caused by 10 degrees are inaudible in practice but I see a lot more aggressive EQ parameters making their way to spinorama scores from time to time.

View attachment 321272

The author's EQ generating code is one of the best and most sophisticated I have seen and comes with no less than 50 optimization paramaters:

spinorama/generate_peqs.py at master · pierreaubert/spinorama

A library to display and compare spinorama (speakers measurements) graphs. - pierreaubert/spinorama

github.com

But that's also my point here!

I would only be concerned if one approaches the suggested EQ to increase the preference score without post validation using their own speakers. Also, the phase changes are small and usually correspond with magnitude response to make the speaker even more linear within the window it's correcting for.

 

[dominikz]

I am a bit concerned about the meaningfulness of the "with EQ" scores of tested speakers. Who decides which parameters should be used to optimize the EQ for each speaker? Does a flatter frequency response mean better sound? What about the phase shifts and delays introduced by these IIR filters, and the changes to the peak energy times of these frequencies?

I understand that the same optimization parameters are used for all speakers to ensure comparability. However, it is possible to auto-EQ every frequency response to be perfectly flat (or on target per se) with the right parameters and method, and this would most probably just sound awful. But there will be cases where some very sharp filters will help improve the response of a certain speaker with phase anomalies in the design, while totally ruining it for another.

The seemingly harmless filters above with a maximum bandwidth of 3 (which I believe to be proportional Q) and a maximum gain of 2.65 dB will still cause more than 10 degrees of phase shifts at many frequency bands (dotted line below). Depending on the actual phase response of the particular speaker, these phase shifts may help or worsen the sound of the speaker. Again, this is a very mild EQ and a few ms of delays caused by 10 degrees are inaudible in practice but I see a lot more aggressive EQ parameters making their way to spinorama scores from time to time.

View attachment 321272

The author's EQ generating code is one of the best and most sophisticated I have seen and comes with no less than 50 optimization paramaters:

spinorama/generate_peqs.py at master · pierreaubert/spinorama

A library to display and compare spinorama (speakers measurements) graphs. - pierreaubert/spinorama

github.com

But that's also my point here!

A few well-known experts on audibility of phase shifts

 

[OCA]

A few well-known experts on audibility of phase shifts

Good reading, thanks. And it confirms that phase shifts are very audible because they cause major SPL changes in the case of stereo listening and in the room environment as opposed to a mono speaker in an anechoic chamber. I have tested more than enough times the improvement in the sound stage precision, centering and height when left and right speakers are phase synched at the MLP which is a logical outcome when left and right speaker delays and inter-driver delays within each speaker are equal.

But my point here is not the phase shifts caused by IIR filters. I am saying that it's possible to improve the score of that particular speaker to 6 or even 7 from 5.6 with enough number of filters and higher Q levels and I am questioning the meaningfullness of evaluating a speaker's sound quality after an EQ applied with some random parameters and use it as a comparison score with another speaker.

 

[GXAlan]

Good reading, thanks. And it confirms that phase shifts are very audible because they cause major SPL changes in the case of stereo listening and in the room environment as opposed to a mono speaker in an anechoic chamber.

+1 and this gets amplified in multichannel environments.

I am questioning the meaningfullness of evaluating a speaker's sound quality after an EQ applied with some random parameters and use it as a comparison score with another speaker.

It's hard to say because the preference score is a great tool for consolidating a lot of information concisely. In doing so, you obviously lose information. I think the patent figure from the Harman preference score is helpful. The highest predicted preference speaker was NOT the most preferred speaker. You can see the one that had a preference score in the mid 6's but actually performed like an 8 and a speaker with a preference score of 5 but sounded like a 3.

 

[sheeple]

You would need to ask @pierre and @Maiky76 to learn about their guidelines they apply to their EQ filters.

It makes perfect sense to apply EQ, as frequency nonlinearity is the strongest factor for loudspeaker preference. You did misread the article @dominikz
provided, picking Wolfgang Klippel’s statement and getting it wrong. The PEQ adjustments of the frequency nonlinearity will not cause any audible issues if the filters are chosen so that the group delay does not exceed frequency dependent thresholds.

Also, any other shortcomings of the speaker will still be reflected in the score. For example, if due to a suboptimal crossover the Power Response showcases a dip around the crossover frequency, it stays and influenced other curves even if linearity of Early Reflections is achieved. So EQ does not distort the rating.

 

[OCA]

You would need to ask @pierre and @Maiky76 to learn about their guidelines they apply to their EQ filters.

I'd be happy to hear their comments on the common guidelines they follow in the EQ process.

if the filters are chosen so that the group delay does not exceed frequency dependent thresholds.

I might have missed but I didn't see any time domain analysis of the response i.e. wavelets, auto-EQ seems to be simply based on the frequency domain response.

if due to a suboptimal crossover the Power Response showcases a dip around the crossover frequency, it stays

Why not include XO phase correction and add it to the score then? Why limit digital correction to IIR filters only?

With all due respect, I still tend to think that it's not optimal to decide a standard auto-EQ process in the freq. domain is the best possible digital correction that can ever be applied to that speaker in any room and classify its sound performance accordingly.

 

[dominikz]

I still tend to believe that it's not optimal to decide a standard auto-EQ process is the best possible digital correction that can ever be applied to that speaker in any room and classify its sound performance accordingly.

Well I can agree that we can't be sure which of the incredibly many EQ approaches are optimal - simply because there is not enough formal research to back it.
We can e.g. optimize for LW flatness, for highest calculated preference score, for PIR smoothness, for phase response smoothness, etc... or for a combination of some (or all) of these. We'd need audibility and preference research for all of these approaches to know which is "best".

However, what we do know from existing audibility and preference research is that people tend to rate loudspeakers with flattish on-axis and smooth directivity higher. We also know that loudspeakers with smooth and even directivity behavior can be EQ-ed to flat on-axis without penalty, since in that case the same filters will be beneficial to responses from all relevant angles. That means several different EQ optimization approaches might result in similar improvements with such loudspeaker.

Having that in mind I'd argue that auto-EQ approaches by @pierre, @Maiky76 and others are very reasonable, at the very least - even if not optimal.

The problem is of course what to do with loudspeakers that have uneven directivity. AutoEQ will still result in score improvements in those cases, but since with such speakers the same EQ filters can't improve the response in all radiation angles, we'd probably need to do deeper perception research to know how to optimize EQ.
Surprisingly (for me), my informal test with one such (poor) loudspeaker indicated optimizing for calculated preference score at the expense of on-axis flatness *may* have some merit.

 

[OCA]

my informal test with one such (poor) loudspeaker indicated optimizing for calculated preference score at the expense of on-axis flatness *may* have some merit.

I just checked the post and I generally refrain to disagree with others in these forums but anyone with some real life experience in DRC will know that IIR filters like a -7dB cutoff at 15kHz (highlighted below) are very unlikely to produce sound improvements in a room. In practice, you don't want to touch above 1kHz unless you have a very good reason and some very robust IR gating method. The treble will sound throttled at best and with a wavelength of 2.2 cm (shorter than the distance between your ears) for 15kHz, the dependence of the benefit of the filter on speaker directivity is also very questionable IMO.

 

[thewas]

I agree that such automatic algorithms still cannot fully replace an expert with his experience on equalisation, but still in most cases they are overall an improvement compared to no EQ at all. Let me quote also Floyd Toole:

"Peaks can be attenuated by EQ, but narrow dips should be left alone - fortunately they are difficult to hear: an absence of sound is less obvious than an excess.
...
The most common flaws in loudspeakers are resonances (which frequently are not visible in room curves) and irregular directivity (which cannot be corrected by equalization). The only solution to both problems is better loudspeakers, the evidence of which is in comprehensive anechoic data.
...
As I said, because loudspeaker transducers are minimum-phase devices one can use electrical parametric EQ to attenuate the mechanical resonances in transducers - using anechoic data of course. So, if you add a hump to an otherwise neutral/resonance free speaker you have added a resonance. This is why it is crucial to pay attention to what "room equalizers" are doing. If they "see" a ripple in a measured curve caused by acoustical interference of direct and reflected sound, and try to flatten it, they may be adding a resonance and degrading a good loudspeaker."

Source and more: https://www.audiosciencereview.com/...ut-room-curve-targets-room-eq-and-more.10950/

 

[IamJF]

"because loudspeaker transducers are minimum-phase devices"

When your speaker has a gentle peak at some frequency the phase will follow. You NEED a IIR EQ cause it corrects frequency response AND phase.

 

[OCA]

Some notes on why you wouldn't want to EQ above 1kHz:

From DRC Designer documentation

Within a room, the sensitivity of the room transfer function to the listening position is roughly dependent on the wavelength involved. This dependence has the side effect that the room correction need to be reduced as the frequency increase, or, seen from the other side, as the wavelength decrease. frequency dependent windowing is implemented with sliding lowpass linear time variant filtering. A tiny fraction of the time-frequency plane gets corrected. This tiny fraction pretty much defines the physical limits where digital room correction is applicable. Above this limit the listening position sensitivity usually becomes so high that even a small displacement of the head from the optimal listening position causes unacceptable results with the appearance of strong audible artifacts.

The graph below illustrates that when the window exponent (exponent used in the frequency dependent window length computation for the band windowing procedure - typical value 1.0) goes below about 0.5, the frequency dependent windowing starts violating the Gabor inequality at least in some small frequency range. Within that range the room transfer function estimation becomes inaccurate and the room correction might be affected by appreciable errors in the evaluation of the room transfer function.

The gray line is the Gabor uncertainty limit which represents the fundamental limit of time and frequency resolution one can extract from a signal:

Gabor uncertainty - SubSurfWiki

subsurfwiki.org

 

[thewas]

Some notes on why you wouldn't want to EQ above 1kHz

That's why Toole (see my above link) and many here recommend and are equalising in that region only based on full anechoic data and not on listening position measurements.

 

[Sokel]

That's why Toole (see my above link) and many here recommend and are equalising in that region only based on full anechoic data and not on listening position measurements.

He also suggest something like to leave alone anything above 500Hz and to correct with a broad low Q self if needed:

Above about 400-500 Hz the "early reflections" curve in the spinorama should be similar to what you have measured. If you have well designed loudspeakers the room curve might have some smallish ripples caused by acoustical interference between and among the direct and reflected sounds - these are not problems to two ears and a brain and equalization is the wrong method of addressing them if they were - that is an acoustics issue. Spatial averaging over several microphone locations tends to smooth the room curve at middle and high frequencies, thereby reducing the likelihood that an auto-EQ algorithm (or a person) might try to "fix" something that can't be fixed, or that doesn't need to be fixed. Remember, any EQ applied to a room curve modifies the direct sound, and it the the direct sound that is a key factor in determining sound quality. If you began with loudspeakers designed to have the desirable smooth and flat on-axis/listening window response, they will be degraded.

What is your favorite house curve

OK. Next question for you folks. As far as I can tell there is no spinorama for the Focal Aria 926's, which happen to be my main speakers. However, Amir did do a spinorama for the Aria 906 standmount speakers. They use the same tweeter and mid driver as the 926, but lack the two woofers as they...

www.audiosciencereview.com

Edit:Right link

 

[OCA]

That's why Toole (see my above link) and many here recommend and are equalising in that region only based on full anechoic data and not on listening position measurements.

Are you saying anechoic responses of speakers need 12dB cuts at 13kHz in this day and age? One doesn't need any test to not buy such designs if they exist!

And back to my main and in fact only point which is being ignored. A fixed parameter auto-EQ process "maybe" should not be a trusted sound score attribute for speaker comparison.

 

[thewas]

Are you saying anechoic responses of speakers need 12dB cuts at 13kHz in this day and age? One doesn't need any test to not buy such designs if they exist!

And back to my main and in fact only point which is being ignored. A fixed parameter auto-EQ process "maybe" should not be a trusted sound score attribute for speaker comparison.

I am afraid you missed my previous fully as it also writes:

I agree that such automatic algorithms still cannot fully replace an expert with his experience on equalisation

 

[OCA]

automatic algorithms still cannot fully replace an expert with his experience on equalisation

I am not arguing on the effectiveness of the automatic algorithm. In fact, I admitted at the very begining that it's one of the best I have seen (I've checked the actual code).

All I am saying is that, a speaker can be digitally corrected in infinite different ways and a certain algo applied to all speakers and than rating them for their sound quality under a "with EQ" parameter might end up unfair. For starters, there's analytical evidence that min phase IIR filters are the most effective below room transient frequency but a mix of IIR & linear phase FIR filters will perform better above that. Similarly, crossover phase linearization improves the sound of almost every speaker and it's also a very applicaple digital correction. Why not score speakers according to this kind of automated filters for instance?

I will stick with the standard "Tone" score of the speakers measured which I believe is a very helpful data we would otherwise never have access to.

 

[ernestcarl]

Are you saying anechoic responses of speakers need 12dB cuts at 13kHz in this day and age? One doesn't need any test to not buy such designs if they exist!

Eh... well, maybe not 12dB at 13kHz exactly, but some speakers and horn loaded transducers will definitely require such "extreme" EQ -- even after some passive crossover network in place.

6HCX51 8Ω - Coaxials - B&C Speakers

www.bcspeakers.com

I use a variant of the 6FHX51 in a slightly modded Fulcrum RX699 (port closed and overdamped -- there is a penalty for that in terms of how loud you can play them after) above my nearfield desk monitor screen which also doubles as a center channel for the couch at the back of my listening room.

The speaker in question sounds quite broken without such extreme EQ. With EQ it sounds very nice side-by-side my KH120s as well as rest of my other studio monitors.

*absent grill and port closed -- the consistency of the frequency response in my listening room post EQ is remarkable above 500 Hz or so:

**dip around 7kHz takes into consideration the directivity index of the speaker

Default configuration with Grill -- slightly increases reflection and diffraction energy loss around ~10 kHz

RX699 – 6” Coaxial Loudspeaker | Fulcrum Acoustic

The RX Series is comprised of high efficiency, high fidelity coaxial loudspeakers that provide output capability typical of a much larger loudspeaker. Their transducer’s high-efficiency and horn-loaded HF compression driver give them a surprisingly high output-to-size ratio, and their broad 90°...

www.fulcrum-acoustic.com

I don't think folks who buy such speakers are idiots. Although... better to know what the hell it is one's doing. Here's a recent installation of the smaller RX599 inside the chandeliers of a church building:

 

[sheeple]

Well,

And back to my main and in fact only point which is being ignored. A fixed parameter auto-EQ process "maybe" should not be a trusted sound score attribute for speaker comparison.

you are free to do it by hand. Optimizing algorithms are a thing, and you would also need to decide if you weight-in the room and reflexions into the speaker transfer function (before room EQ), something engineers do since decades. The decision between On, LW, In-Room remains, and a blind test for comparison is still required.

Please note that you are mixing things up: what applies to room correction is not the same as what we are talking. These EQs are for anechoic response. Its perfectly fine to EQ at any frequency within the speaker passband, that is just the same what an analogue crossover does.

Why not include XO phase correction and add it to the score then? Why limit digital correction to IIR filters only?

How could you EQ the crossover of a multi-way speaker? That is simply not possible, unless you removed the crossover and replace it with another.