What role should listening play in speaker design

[Arindal]

So what is the answer? That we believe his ears and preferences???

The nice thing about free market economies is: Everyone can choose whomever he or she believes. Own ears, AJ´s ears, Dr. Olive´s surveys, measurements, ideology of any flavor...

If I understand AJ correctly, he supports your view on linear response and ideal measurements in the driver development and first stages of loudspeaker design. While relying on his ears when it comes to finding the subjectively best compromise between existing technical imperfections, doing listening tests with a whole lot of different recordings to ensure the subjective results are as consistent as possible.

To me, this sounds like a good approach, but again, no-one has to follow that or buy anything from him.

We are lucky that the preferred on-axis response is flat.

I don´t think we are simply lucky. I would rather assume, the preference for flat on-axis response in tests, is a result of the majority of recordings being mixed or mastered under what is on an average base linear response in studio control rooms. Conditions in studios might vary, but the influence of the room is in majority of cases significantly reduced compared to home listening, and chances are high that at least one individual being involved, like the mastering engineer, will check it on a linear system.

So, I would say the preference for linear response is an averaged result of the ´circle of confusion´, particularly in case purely electronic mixes are presented. No participant is in the position to know how it sounded in the studio and should sound.

Maybe I am biased here, because I have been loosely involved in a process of formulating precise standards to ensure identical monitoring conditions in broadcast control rooms, many many years ago. Just defining an on-axis response, is far from sufficient to define sound quality.

There is zero excuse in producing a speaker with serious deviations from that.

I did not notice AJ calling for intentional ´serious deviations´. Rather calling for a compromise, in case several parameters of relevant measurements cannot meet perfection at the same time. Having witnessed a presentation in his TAD times, explaining about finding a compromise between linear on-axis response and averaged resonse over a broader listening window, to me this sounds reasonable. But again, everyone is free to set own priorities.

 

[ahofer]

Of course this probably won't be the case with his targeted market as otherwise it could become a failure. But we are in ASR and entitled to question the usage of a technology that hasn't till now shown any SOTA products. As said hope to be proven wrong and this will beat every existing TAD, Genelec, Purifi or KEF driver but till then I have my hopefully understandable doubts.

Was going to say, we can judge it however we like, and this site does tend to attract value buyers.

 

[Newman]

So what is the answer? That we believe his ears and preferences???

The nice thing about free market economies is: Everyone can choose whomever he or she believes. Own ears, AJ´s ears, Dr. Olive´s surveys, measurements, ideology of any flavor...

Perhaps you are making an error in comprehension if you think Amir is suggesting the answer involves some limitation on personal freedom. Otherwise, I don't see how you can think the above is a relevant reply.

If I understand AJ correctly, he supports your view on linear response and ideal measurements in the driver development and first stages of loudspeaker design. While relying on his ears when it comes to finding the subjectively best compromise between existing technical imperfections, doing listening tests with a whole lot of different recordings to ensure the subjective results are as consistent as possible.

You are describing in bold the selling of Guru Value. His ears, his listening, almost certainly not blind.

This is exactly the old formula. Excellent driver behaviour at the input stage of speaker design, then magic sauce at the end. Stereophool delight.

To me, this sounds like a good approach, but again, no-one has to follow that or buy anything from him.

It's the evidence-free, trust-me approach. The approach that drove JG Holt to despair. And you reckon that is convincing. And that is your answer to Amir?

I did not notice AJ calling for intentional ´serious deviations´. Rather calling for a compromise, in case several parameters of relevant measurements cannot meet perfection at the same time. Having witnessed a presentation in his TAD times, explaining about finding a compromise between linear on-axis response and averaged resonse over a broader listening window, to me this sounds reasonable.

You are describing in bold the same goal as the formal audio research world of published results, when establishing perceptual models describing the relationships between measurements and listener preference. Which is, AFAICT, not finding these additional magic sauce 'compromises' that the guru designers tell us are baked into their personal designs. At a premium price.

But again, everyone is free to set own priorities.

This is not a question of personal freedoms, that you keep mentioning. They are a given. What makes you think any restrictions are proposed?

Everyone is free to set their own priorities, indeed, principally to prioritise between (a) the findings of formal, peer reviewed, validated, reproducible, controlled science investigations, and (b) guru worship with some variant of audio science stirred in to make it look 'sciency'. Like you said:-

....his point is that while measurements can be accurate (or "true" to a microphone), that does not necessarily translate to "truth" in the sense of sound quality, because the latter is too complex and involving too many compromises compared to a single microphone at a single point.

Yep, it's the old, "science doesn't know enough, yet" guru claim, ignoring the fact prior science has indeed gone well beyond 'true to microphone' measurements and done a lot to correlate key measurements and measurement targets to 'truth in the sense of sound quality'. But instead of advancing the science (firstly by providing valid evidence of the claim that "science doesn't know enough, yet" instead of making an ambit claim, and secondly by advancing the science with new work), instead, take the guru path to insist that there is more (of some unstated thing) and that he has personally found it in his secret magic sauce laboratory. For sale to you, at a premium.

I feel very free.

cheers

 

[Sigberg Audio]

Are we now suggesting loudspeaker designers should not listen to their products before releasing them? I for one would certainly not buy a product from a manufacturer with that approach.

 

[Pearljam5000]

It's like a chef not tasting his own food before serving it

 

[Arindal]

It's like a chef not tasting his own food before serving it

Don´t worry, he does not have to. The ´McDonald´s curve´ is telling him exactly which ingredients and how much salt to put into it. Scientifically proven by surveys with thousands of participants eating tons of cheeseburgers some 4 decades ago.

Fortunately, we are discussing speaker design philosophy here instead, and I want to highlight that this is unrelated to preparing food. So analogies remain completely coincidental and thereby worthless.

 

[amirm]

Are we now suggesting loudspeaker designers should not listen to their products before releasing them? I for one would certainly not buy a product from a manufacturer with that approach.

Every designer does this so it is an odd question.

The point he makes is that some deviations are good because he thinks so. That is fine as long as he conducts controlled listening tests to prove that. Otherwise, it is an excuse in advance of any independent testing that demonstrate those deviations.

Of course we are talking about tonality there. There is plenty to listen for as far as distortion, SPL capability, etc.

 

[amirm]

It's like a chef not tasting his own food before serving it

The "food" in this case is the music itself. We are talking about what plate you serve it on. If that plate adds flavor of its own, then you better have a hell of a justification why that is the case.

 

[dagfinn]

Fact is that we need a standard.

Absolutely. If the pro world leads, the others will follow. In the graphic industry this worked quite well. I'd don't think any pro would touch a non-calibrated monitor. Understanding of ambient lighting, the biology of vision and adaption. We can't escape it, we need to adress it to produce at top level. In visuals it's so much easier, if people are green in the face and the grass is pink, everyone knows something is wrong. If the text is out of focus, it's not OK. NOBODY says "I want a more colorful monitor and some distortion too" - like some people say they prefer over the best and most neutral monitors in audio. As if a perfectly colored rainbow isn't colorful enough, or a perfect square is too square or too much in focus . The understanding of "neutral" is terribly distorted.

Of course, people chose what they like, and some like veneer more than precision. In the pro world, this don't work. What Genelec is doing with GLM and GRADE report is the type of way to go. Transparent speakers and non-modal, even reflection rooms (or-non-reflectant). Plus DSP control and calibration for unbiased neutrality and balance, we simply can't escape biology.

As far as Andrew Jones new speaker, previous results of his works seem to be ok, but we certainly need measurements in order to consider it for pro use. I'm not sure this is meant for pros, though.

 

[Arindal]

Are we now suggesting loudspeaker designers should not listen to their products before releasing them? I for one would certainly not buy a product from a manufacturer with that approach.

Without any irony: If we would have data, which exact parameters/measurements define sound quality for all recordings in question comprehensively, mathematical models which can predict how a speaker interacts with the room, and can ensure that a speaker meets all requirements like target curves with 100% accuracy, I would buy such a product. I see no listening test being mandatory in this case.

In my understanding, human judgment and listening tests in speaker design process, are not meant to bring in personal preferences, a house curve of any kind or subjective ideals. I agree, this should not be the job of a speaker designer. The listening tests in my understanding are necessary to find a balance between several technical aspects, which cannot be all met by the product at the same time, hence demanding a compromise, as well as designing properties for which there is no sufficient prediction model, like interaction between the loudspeaker and a random room.

 

[amirm]

The listening tests in my understanding are necessary to find a balance between several technical aspects, which cannot be all met by the product at the same time, hence demanding a compromise, as well as designing properties for which there is no sufficient prediction model, like interaction between the loudspeaker and a random room.

Only a controlled test will do that. Not listening to random music by yourself and arriving at some conclusion. This is why Revel would not release a speaker without it beating its competition in double blind tests.

You either go through that trouble or just follow the metric of flat on axis and well behaved off-axis. The latter will give you some latitude for decision making but that is it.

 

[Arindal]

Of course we are talking about tonality there.

Agreed. And the whole point which AJ was making in another interview (the one with JA? Don´t remember) is that a compromise made by own ears is mandatory in terms of tonality, if you want to achieve linear reproduction on axis, in the listening window, within the major early reflection windows and the overall directivity index, i.e. 4pi sphere, at the same time.

As far as no speaker is coming even close to meeting all these points at the same time (Kudos to GGNTKT for trying!), what would you suggest as the best method to find a compromise which works for a vast majority of buyers, rooms and recordings?

You either go through that trouble or just follow the metric of flat on axis and well behaved off-axis.

´Well-behaved´ is not defined yet, by comprehensive controlled listening tests. Some people say continuously increasing beaming is ideal, some demand constant directivity, some say listening window, early reflection window, overall d.i., rear hemisphere are important.

To do a meaningful controlled test to define an ideal for any of these, you need to isolate competing properties of a speaker and have them all tested separately in blind A/B comparison. For example speaking about early reflection window tonality, you need one prototype delivering flawless ceiling and floor reflection window, one for side reflection window, one for rear hemisphere, while all must be equally, non-distinguishable linear on axis. Not easy to build.

I have been taking part in a controlled test of identical speakers with identical drivers, crossover freq and anechoic response, showing solely a bunch of differences in the off-axis response of the lower tweeter band (waveguide vs. non-waveguided, in-line vs. coaxial). This alone was a huge endeavor, but results were pretty interesting (and matching what AJ is saying about listening windows and advantages of coaxials for even response over a number of angles).

Only a controlled test will do that.

How would you imagine this taking place in practice? How many prototypes should be built, how many participants in a blind test and how many single experiments does this require?

Not saying the approach is wrong, it sounds pretty logical to me. Just not very practically doable in a speaker design process.

This is why Revel would not release a speaker without it beating its competition in double blind tests.

This is why mastering studios and broadcast institutions, as I am aware of, would not buy such a speaker and install it in their control room and mastering suites, without a double blind test. Interestingly, they seem not to buy Revel and JBL products at all. I wonder, why?

 

[Juhazi]

There is no obligation that loudspeaker designers or manufacturers should follow Amir's advice or "rules".

Optimizing for CTA or other measurement-based indexes is not needed and many customers want speakers that sound in a way they like. Other components in the reproduction chain are transparent and dsp can be used.

This measurement-based trend is very new and narrow. The loyal readers of ASR are a very small portion of consumers!

 

[PristineSound]

The point he makes is that some deviations are good because he thinks so.

Only a controlled test will do that. Not listening to random music by yourself and arriving at some conclusion.

Any speaker company does this, I absolutely will not even give it a time or the day. Not only that, I will actively go around letting other potential customers that I come across know that the company has decided for you what EQ tweaks sounds good.

There are some acoustic engineers/speaker designers who will mess around with the FR curve to anticipate how the speaker design will interact with the room, but with the goal to achieve a tonally balanced in room response. Even that, I'm apprehensive about.

Those who mess around with FR rooted in subjectivism is very anti science to me. No thanks, I'm not interested in your so called "art".

In my opinion, the speaker should be ruler flat on axis with controlled directivity, then let the listener mess around with the EQ as they choose. Otherwise the circle of confusion gets more and more confusing.

 

[Arindal]

In my opinion, the speaker should be ruler flat on axis with controlled directivity

Ruler-flat at which particular axis?

And what really defines ´controlled directivity´ I your understanding, in which frequency range is it important, what is the target d.i.? Does it translate to constant directivity, alternating d., continuously increasing or decreasing directivity? Which aspects of off-axis behavior should be ´controlled´, the ceiling bounce window, floor bounce, side-wall reflections, rear hemisphere, overall hemisphere, main listening window for one listener (sitting at which horizontal angle exactly?) or several? What is the acceptable tolerance of deviations?

My answer: ruler-flat constant directivity for all the aforementioned windows at the same time, from 300 to 8,000 Hz, 1/3 of an octave smoothing, d.i. depending on RT60 and desired listening distance, is sufficient. Once this is achieved, I agree we do not need listening tests for judging tonality and finding a compromise anymore.

Otherwise the circle of confusion gets more and more confusing.

In my understanding, the circle of confusion ist at maximum play if one does controlled tests involving recordings unknown to the participants, solely existing as an electronic piece of art being produced by trusting monitors/headphones.

If you want to break at least partially this circle of confusion, you have to have an acoustic reference, and listeners who can judge differences because they know the original. I found recording engineers and broadcast mixdown engineers being into classical music and having a habit of listening to the original event, to be the most reliable sources.

This measurement-based trend is very new and narrow.

I agree that it is narrow, particularly with people being in the market for up-price hi-fi gear.

I have my doubts about ´new´. I am aware of the first test publications claiming everything sound-quality-related can be measured and determined, hence publishing a ton of hi-fi measurements, dating back to 1962. Would rather asses that somewhen, maybe in the 1980s, a huge fraction of the market abandoned the idea of measurement determination as a whole, with many plunging into the opposite extreme of unhinged subjectivism and esoteric beliefs (which I am maximum skeptical of).

The interesting question is: why?

Either way, thanks @Juhazi for sharing this old interview of AJ pointing out he is a physicist and was head of R&D at KEF, clarifying quite some questions on his stance on measurements and at which stage they are useful.

 

[SSS]

Every designer does this so it is an odd question.

The point he makes is that some deviations are good because he thinks so. That is fine as long as he conducts controlled listening tests to prove that. Otherwise, it is an excuse in advance of any independent testing that demonstrate those deviations.

Of course we are talking about tonality there. There is plenty to listen for as far as distortion, SPL capability, etc.

What do you understand as a controlled listening test?

 

[Sokel]

What do you understand as a controlled listening test?

You know the amusement park's tea cups that spin?
You would need something similar for perfect accuracy.

Anything else in a room (even the very speakers next to each other) will interact differently without such a system or else it would take a long time, which is a no-go.

Edit: and that, without complexities like different acoustic center (that you would adjust listening height at normal circumstances at MLP) , etc .

 

[TimVG]

Are we now suggesting loudspeaker designers should not listen to their products before releasing them? I for one would certainly not buy a product from a manufacturer with that approach.

It would depend on how good the designer is with interpreting all the various measurements

 

[Marin]

Ruler-flat at which particular axis?

And what really defines ´controlled directivity´ I your understanding, in which frequency range is it important, what is the target d.i.? Does it translate to constant directivity, alternating d., continuously increasing or decreasing directivity? Which aspects of off-axis behavior should be ´controlled´, the ceiling bounce window, floor bounce, side-wall reflections, rear hemisphere, overall hemisphere, main listening window for one listener (sitting at which horizontal angle exactly?) or several? What is the acceptable tolerance of deviations?

My answer: ruler-flat constant directivity for all the aforementioned windows at the same time, from 300 to 8,000 Hz, 1/3 of an octave smoothing, d.i. depending on RT60 and desired listening distance, is sufficient. Once this is achieved, I agree we do not need listening tests for judging tonality and finding a compromise anymore.



In my understanding, the circle of confusion ist at maximum play if one does controlled tests involving recordings unknown to the participants, solely existing as an electronic piece of art being produced by trusting monitors/headphones.

If you want to break at least partially this circle of confusion, you have to have an acoustic reference, and listeners who can judge differences because they know the original. I found recording engineers and broadcast mixdown engineers being into classical music and having a habit of listening to the original event, to be the most reliable sources.



I agree that it is narrow, particularly with people being in the market for up-price hi-fi gear.

I have my doubts about ´new´. I am aware of the first test publications claiming everything sound-quality-related can be measured and determined, hence publishing a ton of hi-fi measurements, dating back to 1962. Would rather asses that somewhen, maybe in the 1980s, a huge fraction of the market abandoned the idea of measurement determination as a whole, with many plunging into the opposite extreme of unhinged subjectivism and esoteric beliefs (which I am maximum skeptical of).

The interesting question is: why?

Either way, thanks @Juhazi for sharing this old interview of AJ pointing out he is a physicist and was head of R&D at KEF, clarifying quite some questions on his stance on measurements and at which stage they are useful.

Yes this interview reveals a lot about him and his beginnings as an audio engineer...
It's funny that he now supposedly doesn't believe in measurements, but he started his career at KEF creating tools for measuring speakers.
Also people that see Andrew listening to speaker in room with an inductor in one hand and a capacitor in the other are dead wrong.
Likewise he admits that he did not hear the finished product until the show.

 

[Arindal]

It would depend on how good the designer is with interpreting all the various measurements

Could you name an example of such an experienced , trustworthy loudspeaker designer, pls?

And how do you think it is possible to judge from outside, without involving own listening tests, how good a designer is with interpreting? Would that not qualify as ´believing in Guru value´ and bringing in ´trust-me-approach´ and audiophool beliefs, as some participants in this thread have put it?

It's funny that he now supposedly doesn't believe in measurements, but he started his career at KEF creating tools for measuring speakers.

Funnily, I don´t find that surprising, given the fact he several times highlights the importance and number of measurements in driver development and early stages of loudspeaker design. He just seemingly denies the importance of measurements as the last and only institution to make decisions on compromises in final products.

Did you consider the possibility of him coming to the latter conclusion, *because* he had been creating measurement tools and worked with a ´measurements first´ company early in his career?