Neumann KH 80 DSP Monitor Review

[amirm]

Since when? Not in Toole's days!

Post Dr. Toole's days. I did ton of due diligence before buying this system, including checking with Dr. Toole. Please don't question me on what I know. I know what I know!

 

[amirm]

Why would someone do listening comparison tests on a desk without EQ when it is known that the peak around 150 Hz is dentrimental for neutral tonality?

Why would you listen to any speaker indoor since that massively changes the low frequency response? Which is detrimental to sound quality?

Why the heck do you think I am measuring speakers using anechoic techniques? It is so that we have objective, room independent analysis of the speaker sound.

Plenty of people slap these things on their desk and use them that way. That is by far more common than the people who use any type of EQ, much less that 150 Hz correction based on potentially faulty evaluations years ago.

 

[amirm]

I think you misunderstood my quote there, I said that tonality is the most important to preference when it comes to blind testing a la Harman, but when both "A" and "B" are there almost perfect than other important factors like IMD become the decisive factor.

Current research doesn't back your point of view. As I mentioned read the AES paper, "The Correlation Between Distortion Audibility and Listener Preference in Headphones"

Steve Temme , Sean E. Olive , Steve Tatarunis , Todd Welti , and Elisabeth McMullin, Listen, Inc., Boston, MA, 02118, USA
Harman International Industries, USA

The study involved equalizing four headphone to the same frequency response and then emulating that on a neutral headphone. Result was that distortion was not a predictor of preference even in that situation. With respect to your IMD comment, here is the paper:

This is to be expected because these traditional methods of distortion are not psychoacoustically aware. This is why all of my electronics measurements include full spectrum of the distortion so that we can apply psychoacoustics to them. A blunt distortion metric of IMD/THD, etc. is not that useful other than something to chew on when you are bored.

 

[amirm]

Bass modes are one thing, but at desktop placement the higher 150 Hz peak is the most serious tonal offender in my experience of such setups.

And opposite of my experience. I suggest having a blind test done to make sure you can even hear that effect let alone have it be the worst offender. Room modes substantially destroy fidelity of music. No way this narrowband peak has the same effect. Floor reflections which is an analog of this has been shown to not be as relevant as one may think. See Dr. Toole and my response a bit later: https://www.audiosciencereview.com/...y-without-measurement.7127/page-3#post-162809

 

[amirm]

And therein lies some of the issues with speaker measurements. Is it designed to be placed against a wall or on a desk or on stands? With a sub or without? Does it work better in this room or that? Near field, mid field, or far field? Perfectly flat or Harman flat? Can a DSP make up for all of its flaws? How loud can it go? How well will it integrate with a sub? So many things play into it all.

There is no issue in the larger context of how to optimize sound reproduction in a room. Speaker response is a key and super important factor. It is not however the only factor as the room obliterates the speaker frequency response below a few hundred hertz. Subs are a tool to solve that puzzle. Once you have an integrated picture and data for all of this, you can optimize to get fantastic sound. Sit there with no measurements of the speaker and you have no idea what is going on as a single microphone measurements product improper data above transition frequencies.

 

[bobbooo]

Yeah, but with this, the smoothness is drastically lower because it doesn't follow the target slope, it's almost flat for the in-room.
I'm trying to see if I can get smoothness to only look at smoothness regardless of slope, but I'm not having luck. Any alterations to target slope have no effect on the final score.

Surely the fact that the smoothness variable SM (and so final score) doesn't change if the target slope is changed, tells us that following the target slope is not a factor influencing SM? Which is as it should be - SM quantifying only smoothness, and SL only slope. It seems like SL being used in the formula for SM just cancels out somewhere in the spreadsheet maths. But then it does seem odd that it's included in the SM formula in the first place. I think we need clarification from Sean Olive on this.

 

[MZKM]

Surely the fact that the smoothness variable SM (and so final score) doesn't change if the target slope is changed, tells us that following the target slope is not a factor influencing SM? Which is as it should be - SM quantifying only smoothness, and SL only slope. It seems like SL being used in the formula for SM just cancels out somewhere in the spreadsheet maths. But then it does seem odd that it's included in the SM formula in the first place. I think we need clarification from Sean Olive on this.

Yeah, any alteration to the target slope makes 0 difference in the final smoothness score (R2); so I’m not sure why it’s even relevant to the calculation, unless there is something I‘m missing, such as why the summation goes from 12kHz to 100Hz, when typical summation is done from min to max.

I found a Reddit discussion of someone asking about the same thing (near 0 slope giving them worse results), and the comments just state that it’s irrelevant.

I won’t know until Amir measures more speakers, if we get one with a flat PIR but has little variation, then we’ll know that the 305’s score is correct and that R2 is just really aggressive.

 

[neutralguy]

This is to be expected because these traditional methods of distortion are not psychoacoustically aware. This is why all of my electronics measurements include full spectrum of the distortion so that we can apply psychoacoustics to them.

How do we apply psychoacoustically aware analyses on them? If you have in mind perceptual metrics such as Rnonlin, @MZKM: would it be possible to write a program to compute them as well? They could apply to electronics and speakers, in theory.

 

[pozz]

How do we apply psychoacoustically aware analyses on them? If you have in mind perceptual metrics such as Rnonlin, @MZKM: would it be possible to write a program to compute them as well? They could apply to electronics and speakers, in theory.

It's too complex and system/listener dependant, and won't fit on a single graph. It's far easier to read Toole's book and learn to interpret the spinoramas.

Sorry. No easy way to knowledge.

 

[bobbooo]

Yeah, any alteration to the target slope makes 0 difference in the final smoothness score (R2); so I’m not sure why it’s even relevant to the calculation, unless there is something I‘m missing, such as why the summation goes from 12kHz to 100Hz, when typical summation is done from min to max.

I found a Reddit discussion of someone asking about the same thing (near 0 slope giving them worse results), and the comments just state that it’s irrelevant.

I won’t know until Amir measures more speakers, if we get one with a flat PIR but has little variation, then we’ll know that the 305’s score is correct and that R2 is just really aggressive.

Olive writes the bounds of all the summations like that, with upper and lower bounds inverted. I wouldn't read anything into that, the summation is commutative anyway.

It does seem like something has been lost in translation between Olive's description of SM and the spreadsheet formula, not sure what though.

 

[neutralguy]

It's too complex and system/listener dependant, and won't fit on a single graph. It's far easier to read Toole's book and learn to interpret the spinoramas.

Sorry. No easy way to knowledge.

No, nonlinear distortion is a completely different thing than Toole's work and spinoramas. I think you can catch up by reading some of the previous posts in this thread.

 

[pozz]

No, nonlinear distortion is a completely different thing than Toole's work and spinoramas. I think you can catch up by reading some of the previous posts in this thread.

Definitely not.

There is no correlation between nonlinear distortion and listener preference, or even listener ability to detect, in the literature Toole reviewed. His final opinion on the matter, and I quote, that "the only meaningful number is zero" given that everything above that is questionable.

Toole mentions for example the work by Geddes and Lee, their metric having an established correlation with listening tests, but notes that unfortunately that hasn't been fully developed as yet and that, overall, distortion is far less important than the data he established through listening tests and anechoic measurements.

 

[neutralguy]

There is no correlation between nonlinear distortion and listener preference, or even listener ability to detect, in the literature Toole reviewed.

You may have missed my link for Rnonlin: http://www.aes.org/e-lib/online/browse.cfm?elib=13013
This metric came after Toole's work and his conclusions. Most previous conclusions about nonlinear distortion were drawn based on THD and IMD which indeed correlate poorly with perception.

"A new method predicts the perceived quality of nonlinearly distorted signals based on the outputs of an array of gammatone filters in response to the original signal and the distorted signal. [...] gives a perceptually relevant measure of distortion called Rnonlin, which can be used to predict subjective ratings."

 

[pozz]

You may have missed my link for Rnonlin: http://www.aes.org/e-lib/online/browse.cfm?elib=13013
This metric came after Toole's work and his conclusions. Most previous conclusions about nonlinear distortion were drawn based on THD and IMD which indeed correlate poorly with perception.

"A new method predicts the perceived quality of nonlinearly distorted signals based on the outputs of an array of gammatone filters in response to the original signal and the distorted signal. [...] gives a perceptually relevant measure of distortion called Rnonlin, which can be used to predict subjective ratings."

Sorry, I did miss that.

I've downloaded the paper and I'll give it a read.

 

[amirm]

You may have missed my link for Rnonlin: http://www.aes.org/e-lib/online/browse.cfm?elib=13013
This metric came after Toole's work and his conclusions.

Dr. Toole's book very much post dates that paper.

From my read of it, the formula requires knowing what the non-linear transfer function of the system is, to compute its coefficient. As a practical matter, such a function does not exist.

 

[pozz]

@audiotruth

Rnonlin is based on nonlinearity of contrived digital signals and clipping (which they call "artificial distortions") and nonlinearity of small drivers fed in the following way to produce "real distortions":

Work was supported by Nokia, so it makes sense why they used just these kinds of drivers.

I think this paper isn't relevant for full-range speakers, but only for small drivers which are expected to be (over)driven as part of their operation, as in cellphones and laptops and so on, the object being to overdrive them in a controlled manner which produces the least objectionable kind of distortion possible.

Correlations were higher than 0.9 for both music and speech, for what it's worth. The method is very sound, using ERBs to define filter bands for subjective testing.

 

[neutralguy]

Dr. Toole's book very much post dates that paper.

The book is published recently but it's based primarily on research from the 80's and 90's. The book cites a related work by the same team and other more recent works on nonlinear distortion and says these "new investigations [...] provide additional useful perspectives". Not to speak for him, but it doesn't sound like he has a conclusion about these works.

 

[neutralguy]

I think this paper isn't relevant for full-range speakers, but only for small drivers which are expected to be (over)driven as part of their operation, as in cellphones and laptops and so on, the object being to overdrive them in a controlled manner which produces the least objectionable kind of distortion possible.

Thanks for the reading. I don't imagine they're perfect for our purposes either, but I would think it, or other recent metrics, has at least the potential to give us a better signal than THD/IMD. The research also tells us that freq response may not be the only relevant audible measurement.

And in context, this is only if we want to go further than frequency response, which I agree is a far bigger issue for most speakers.

 

[pozz]

Thanks for the reading. I don't imagine they're perfect for our purposes either, but I would think it, or other recent metrics, has at least the potential to give us a better signal than THD/IMD. The research also tells us that freq response may not be the only relevant audible measurement.

And in context, this is only if we want to go further than frequency response, which I agree is a far bigger issue for most speakers.

It's a start for sure.

 

[BYRTT]

Microphone curve based the five numbers variation in high frequencies amirm shared including a 3,5Hz 1st order high pass knee based datasheet of MK 255, on a 50dB scale it doesn't mean much calibrate for response or not it has beatiful low end reach where NFS seems to run over anechoic data in resolution, for @napilopez puzzle of variations in the various KH 80 curves added one standard UMIK-1 into graph .