Headphones and the Harman target curve

[Robbo99999]

People seem to forget that the Harman research actually specifies a range. "Expert listeners" prefer less bass and the uninitiated prefer more, so the Harman target is a middle ground that makes no one happy.

The good thing is that the bass level is the easiest to tweak in a general manner, just add a 105Hz Low Shelf Q0.7 filter and then set it to +3dB or -3dB or whatever amount of Gain suits your preference. I prefer bass bang on the Harman Target, so it makes me happy. Although I do like the bass to start rolling away from the Harman Curve at around 35Hz & below, as I find that improves clarity through the whole frequency range.

 

[Yorkshire Mouth]

Wow can’t believe I missed this. It makes absolutely no sense.
The measurement of listener preferences as part of an experiment can absolutely be considered “science.”

Okay, firstly my apologies for the delay. Now I've already said this, but it needs repeating, as it appears some have missed it.

We have accuracy. And we have preference. A preference for the inaccurate can be measured scientifically, but that doesn't make the end preference 'scientific'.

Let's take what has become a lazy stereotype. Lazy? Yes. A stereotype? Yes. But also useful.

A kid in their late teens in a car with a massive, unmusical subwoofer in the boot (US trunk).

It's what he prefers. It's his preference. You can question subwoofer-in-car-loving people, ask them about their preference in a scientific way, and scientifically weight their answers using all the tools available to the psephologist. That doesn't make their preference 'scientifically proven'.

The Harman target is the result of some excellent, and extensive research, but all of it is ultimately filtered through the preference of the listeners, who apparently took a 'great speaker in a great room', and decided it'd sound better if they turned the bass knob up.

And there's no getting away from that.

Measuring their desire to do that can be done scientifically, in that their preferences were scientifically measured. But I don't feel that's what Audio Science Review does. And I'll use this example again. Amir tests a DAC or amp, and if the frequency response is ruler-flat, he praises it.

Now let's suppose Topping release 2 sister versions of a new DAC. They're identical, except one has ruler-flat frequency response, the other has 6 dB boosted base.

So before [ub;ishing his review(s), Amir asks Topping why this is. And they say they tested the DACs with listeners, on 'great speakers in a great room' (indeed, the same speakers and room used for Harman), and people preferred the bassy DAC to the ruler-flat DAC.

How should Amir respond in his review(s)? Which of the 2 is more 'scientific'?

I'm not saying the Harman curve is wrong. For all sorts of reasons I suspect it's right, or close. I'm just saying you can't say it's 'scientifically correct' in anything other than a measurement of preference, as opposed to accuracy.

To be fair, I'm not sure it's supposed to be anything else, but appears to have become so in some people's minds.

 

[preload]

Okay, firstly my apologies for the delay. Now I've already said this, but it needs repeating, as it appears some have missed it.

We have accuracy. And we have preference. A preference for the inaccurate can be measured scientifically, but that doesn't make the end preference 'scientific'.

Let's take what has become a lazy stereotype. Lazy? Yes. A stereotype? Yes. But also useful.

A kid in their late teens in a car with a massive, unmusical subwoofer in the boot (US trunk).

It's what he prefers. It's his preference. You can question subwoofer-in-car-loving people, ask them about their preference in a scientific way, and scientifically weight their answers using all the tools available to the psephologist. That doesn't make their preference 'scientifically proven'.

The Harman target is the result of some excellent, and extensive research, but all of it is ultimately filtered through the preference of the listeners, who apparently took a 'great speaker in a great room', and decided it'd sound better if they turned the bass knob up.

And there's no getting away from that.

Measuring their desire to do that can be done scientifically, in that their preferences were scientifically measured. But I don't feel that's what Audio Science Review does. And I'll use this example again. Amir tests a DAC or amp, and if the frequency response is ruler-flat, he praises it.

Now let's suppose Topping release 2 sister versions of a new DAC. They're identical, except one has ruler-flat frequency response, the other has 6 dB boosted base.

So before [ub;ishing his review(s), Amir asks Topping why this is. And they say they tested the DACs with listeners, on 'great speakers in a great room' (indeed, the same speakers and room used for Harman), and people preferred the bassy DAC to the ruler-flat DAC.

How should Amir respond in his review(s)? Which of the 2 is more 'scientific'?

I'm not saying the Harman curve is wrong. For all sorts of reasons I suspect it's right, or close. I'm just saying you can't say it's 'scientifically correct' in anything other than a measurement of preference, as opposed to accuracy.

To be fair, I'm not sure it's supposed to be anything else, but appears to have become so in some people's minds.

There's probably a nicer way of saying this, but the answers to all of your rhetorical examples are pretty obvious for anyone with at least a basic (gonna say high school) understanding of what science is. The short answer is that taking a non-representative subject with n=1 (ie the teenager who likes distorted bass) and generalizing observations made to the rest of the population is a very bad methodology. However, taking a sample of 1,000 teenagers and seeking their preferences under controlled conditions, and generalizing your conclusion to teenagers, would be more authoritative. Both are still considered a scientific experiment, but your example would be considered a very poor methodology.

The problem I feel is that you appear to be confusing the terms "preference" and "accuracy." They are not necessarily the same. Preference experiments measure preference directly. To measure PERCEPTION of accuracy, you would need an entirely different experiment, such as asking human subjects to listen to music in the studio then on a comparison set of loudspeakers in a room. And you can immediately see how cumbersome thst would be and it's pretty clear why Harman decided to measure preference, which btw, I think is a more important outcome anyway.

But to say that measuring preferences under controlled conditions is not "scientific" is just asinine. As someone said on another forum, you should educate yourself before trying to educate others (or something like that).

 

[Yorkshire Mouth]

There's probably a nicer way of saying this, but the answers to all of your rhetorical examples are pretty obvious for anyone with at least a basic (gonna say high school) understanding of what science is. The short answer is that taking a non-representative subject with n=1 (ie the teenager who likes distorted bass) and generalizing observations made to the rest of the population is a very bad methodology. However, taking a sample of 1,000 teenagers and seeking their preferences under controlled conditions, and generalizing your conclusion to teenagers, would be more authoritative. Both are still considered a scientific experiment, but your example would be considered a very poor methodology.

The problem I feel is that you appear to be confusing the terms "preference" and "accuracy." They are not necessarily the same. Preference experiments measure preference directly. To measure PERCEPTION of accuracy, you would need an entirely different experiment, such as asking human subjects to listen to music in the studio then on a comparison set of loudspeakers in a room. And you can immediately see how cumbersome thst would be and it's pretty clear why Harman decided to measure preference, which btw, I think is a more important outcome anyway.

But to say that measuring preferences under controlled conditions is not "scientific" is just asinine. As someone said on another forum, you should educate yourself before trying to educate others (or something like that).

Well, I'll ignore the rudeness.

When you say "The problem I feel is that you appear to be confusing the terms "preference" and "accuracy. They are not necessarily the same." I'm not confusing them at all, nor saying they're the same. I'm saying the opposite, that people shouldn't confuse the two, because they're not the same.

Then when you say "But to say that measuring preferences under controlled conditions is not "scientific" is just asinine," again completely misses the point. I'm saying (and have been abundantly clear) that the measurements of preference CAN and indeed HAVE BEEN (in this research) scientifically carried out, but that that's only a scientific measurement of preference, not of accuracy. Which comes back to your last point, where you say I'm saying the two are the same, when I'm saying the opposite.

Please refrain from using terms like 'you don't have a high school understanding of science', and 'asinine'. They're not in the spirit of serious debate, which should about issues, not people. They provide far more heat than light, and don't advance the discussion.

So going back to the start of your post, yes there's a nicer way of saying what you've said. Simply don't use insulting terms and name-calling.

Discuss the issues, not individuals. Play the ball, not the man.

 

[preload]

Well, I'll ignore the rudeness.

When you say "The problem I feel is that you appear to be confusing the terms "preference" and "accuracy. They are not necessarily the same." I'm not confusing them at all, nor saying they're the same. I'm saying the opposite, that people shouldn't confuse the two, because they're not the same.

Then when you say "But to say that measuring preferences under controlled conditions is not "scientific" is just asinine," again completely misses the point. I'm saying (and have been abundantly clear) that the measurements of preference CAN and indeed HAVE BEEN (in this research) scientifically carried out, but that that's only a scientific measurement of preference, not of accuracy. Which comes back to your last point, where you say I'm saying the two are the same, when I'm saying the opposite.

Please refrain from using terms like 'you don't have a high school understanding of science', and 'asinine'. They're not in the spirit of serious debate, which should about issues, not people. They provide far more heat than light, and don't advance the discussion.

So going back to the start of your post, yes there's a nicer way of saying what you've said. Simply don't use insulting terms and name-calling.

Discuss the issues, not individuals. Play the ball, not the man.

Thank you for affirming that the measurement of listener preferences can be regarded as scientific.

If you feel insulted by what I wrote, you might want to also ask yourself how insulted the people who designed, conducted, and published the preference research should feel after you dismissed their groundbreaking work as not scientific.

 

[Yorkshire Mouth]

Thank you for affirming that the measurement of listener preferences can be regarded as scientific.

If you feel insulted by what I wrote, you might want to also ask yourself how insulted the people who designed, conducted, and published the preference research should feel after you dismissed their groundbreaking work as not scientific.

To be clear, I've not said 'the work' is unscientific, and I've praised those involved for what they've done. I just think that others here have used the preference curve as a scientifically-proven & accurate benchmark for sound reproduction, akin to having ruler-flat frequency response from a DAC; something I don't believe was the initial intent of the research.

So I don't feel I've insulted them. The fact that you feel that, in your opinion, I've insulted them, so this justifies you insulting me, is rather troubling in itself. You've effectively just said "Okay, I'm being insulting, but you started it."

I'd hope we can rise above that.

 

[preload]

To be clear, I've not said 'the work' is unscientific, and I've praised those involved for what they've done. I just think that others here have used the preference curve as a scientifically-proven & accurate benchmark for sound reproduction, akin to having ruler-flat frequency response from a DAC; something I don't believe was the initial intent of the research.

So I don't feel I've insulted them. The fact that you feel that, in your opinion, I've insulted them, so this justifies you insulting me, is rather troubling in itself. You've effectively just said "Okay, I'm being insulting, but you started it."

I'd hope we can rise above that.

This is what you wrote:
"I think we can put to bed the ‘science’ bit.
Preference isn’t science."

So to be clear, are you now saying that preference IS science?

Btw you may be confusing "insult" from "observation." Perhaps consider whether there's any merit to what was stated about what your comment implies about your understanding of science.

 

[JJB70]

Any headphone preference is inherently subjective for the simple reason it is something you wear (or stick in your ears in the case of IEMs). The most important quality of a headphone is comfort as if you can't stand wearing the thing then how good it sounds is irrelevant. And comfort is a very personal thing. I love my Etymotic ER4SR but it is easy to understand why some hate the deep insertion fit. I really liked the One More triple driver headphone sound quality but couldn't buy them as I found them to be really uncomfortable.

 

[moosso]

The harman curve is more like a reference than a destination to me, for example I like flat and relatively light bass I will bring 20~200Hz down and flatten it, then make the mid and treble close to harman curve. Sometime I think the vocal is a little behind I will bring 1khz~3khz 1db above the target so vocals would sound more "forward" but still nature.
The hardest part is the frequence response measurement may not like what I actually hear, I'm not the dummy head and my ear isn't a microphone, some narrow dip and peaks may shift to other location, I have to use swept sine wave to confirm if I done right.
After I learned how to use the parametric EQ I tend to buy comfortable and lightweight (better <300g) one, like 5XX and 6XX openbacks from Senny, the default sound signature become less important to me.
But I agree someone said harman curve is boring, some headphone with unique sound signature did sound more "funny", make it sound the same as other headphones is like taking it's talent away.

 

[jae]

Okay, firstly my apologies for the delay. Now I've already said this, but it needs repeating, as it appears some have missed it.

We have accuracy. And we have preference. A preference for the inaccurate can be measured scientifically, but that doesn't make the end preference 'scientific'.

Let's take what has become a lazy stereotype. Lazy? Yes. A stereotype? Yes. But also useful.

A kid in their late teens in a car with a massive, unmusical subwoofer in the boot (US trunk).

It's what he prefers. It's his preference. You can question subwoofer-in-car-loving people, ask them about their preference in a scientific way, and scientifically weight their answers using all the tools available to the psephologist. That doesn't make their preference 'scientifically proven'.

The Harman target is the result of some excellent, and extensive research, but all of it is ultimately filtered through the preference of the listeners, who apparently took a 'great speaker in a great room', and decided it'd sound better if they turned the bass knob up.

And there's no getting away from that.

Measuring their desire to do that can be done scientifically, in that their preferences were scientifically measured. But I don't feel that's what Audio Science Review does. And I'll use this example again. Amir tests a DAC or amp, and if the frequency response is ruler-flat, he praises it.

Now let's suppose Topping release 2 sister versions of a new DAC. They're identical, except one has ruler-flat frequency response, the other has 6 dB boosted base.

So before [ub;ishing his review(s), Amir asks Topping why this is. And they say they tested the DACs with listeners, on 'great speakers in a great room' (indeed, the same speakers and room used for Harman), and people preferred the bassy DAC to the ruler-flat DAC.

How should Amir respond in his review(s)? Which of the 2 is more 'scientific'?

I'm not saying the Harman curve is wrong. For all sorts of reasons I suspect it's right, or close. I'm just saying you can't say it's 'scientifically correct' in anything other than a measurement of preference, as opposed to accuracy.

To be fair, I'm not sure it's supposed to be anything else, but appears to have become so in some people's minds.

"I'm just saying you can't say it's 'scientifically correct' in anything other than a measurement of preference, as opposed to accuracy"

No need to mince words or have any confusion. It's a scientifically accurate measurement of FR preference in a population in the conditions in which they were performed.

If a clinical trial were employed to determine the efficacy, side effects, or magnitude of a therapeutic dose of a medication in a population, would it also be "unscientific" by these standards if they are based on subjective reports? If such a study determined that the ideal dosage was 50 milligrams per day (and factors like height/weight/age negligibly impacted therapeutic outcomes), would you really not consider this dose "scientifically correct"?

Unless you are going to make some argument on their methodology or claim their samples were exclusionary or the size were not adequate, then there is really no argument on the scientific veracity or accuracy of their paper. They scientifically determined what it was they set out to do and I don't doubt it is reproducible under their conditions.

One should be amazed that despite how many different settings one can enjoy music, how many different genres people enjoy, how many different languages they speak, the volumes they listen at, how many different ear anatomies, how different humans are in general and how different humans claim to be from one another, that preference amongst multiple cohorts and large populations seem to hold invariably between a limited range of +/- a few dB with healthy listeners.

Fundamentally a DAC is a device that is meant to accurately convert a signal from the digital to analogue domain, which we can measure. An amplifier is a device that is meant to increase the power of a signal with ideally zero/minimal loss, which we can measure. Amir ultimately reviews products, with a heavier weight on the measurements (as this audio science is what the forum is about) but still holistically to some degree. I would think Amir more or less goes into his reviews with these assumptions of what a DAC/AMP is and is trying to accomplish. If an dac or amp was intentionally designed to have some sort of FR colouration, or that it added a specific amount of Nth harmonic distortion and these were marketed "features", I don't doubt he would give it praise or recommend it if it fulfilled what it set out to do from a scientific standpoint and it could be proven with a measurement. Short of that, why wouldn't all electronics be treated as the devices they are claiming to be and not some sort of effects box?

 

[bluefuzz]

We have accuracy. And we have preference.

In the context of headphones it amounts to the same thing.

We can measure electronics to arbitrary levels of accuracy. We can measure sound with microphones implanted in peoples ear canal or build dummy heads to approximate an typical HRTF. But what we cannot do is measure subjective experience. Once a soundwave has passed through the eardrum and the complex machinery of the inner ear it becomes spike trains in the auditory systems of the brain and ultimately … what? Music, words, enjoyment, i.e. subjective experience. But we have as yet no way to measure how that experience relates to the outside world and in this case the perceived frequency response of a headphone. We can put electrodes on peoples heads and watch squiggles on an EEG or measure bloodflow to various parts of the brain but we don’t really know what these things mean in relation to our experience. We cannot quantify qualia.

So all we can do is to actually ask people whether the sounds received via a headphone sound right to them. ‘A bit more bass?’ … ‘a bit less midrange?’ … Ultimately these questions about people’s preference is the only possible way to establish any form of ‘accuracy' in headphone reproduction – and this of course is what Harman have done ...

 

[DeLub]

Ultimately these questions about people’s preference is the only possible way to establish any form of ‘accuracy' in headphone reproduction – and this of course is what Harman have done ...

So you're saying that the preferred bass bump is accurate? I can understand a preference for extra bass, but this is not necessarily accurate. To me it is more like "I love the color green, let's bump up green in my tv picture", which most of us certainly won't do. Similarly, when I listen to e.g. piano music, the pianist has put great care in balancing the volume between his left en right hands. I'd like to hear it as he/she intended it, I do not want to hear extra bass, just because that's the way I like it.

I have set my tv to a white balance of 6500K, because that's the agreed standard, this ensures I see the picture the way it was intended. A target curve in sound reproduction should achieve the same. Unfortunately there is no such standard and we have to deal with the circle of confusion. This makes it all way more complicated. And although I agree we can investigate what the target curve should be by asking people what they think of it, there is no reason to include an obviously preferred bass bump in a target curve. Instead we should not ask people what they prefer, but what they think sounds more "natural", real life like.

 

[preload]

So you're saying that the preferred bass bump is accurate? I can understand a preference for extra bass, but this is not necessarily accurate. To me it is more like "I love the color green, let's bump up green in my tv picture", which most of us certainly won't do. Similarly, when I listen to e.g. piano music, the pianist has put great care in balancing the volume between his left en right hands. I'd like to hear it as he/she intended it, I do not want to hear extra bass, just because that's the way I like it.

I have set my tv to a white balance of 6500K, because that's the agreed standard, this ensures I see the picture the way it was intended. A target curve in sound reproduction should achieve the same. Unfortunately there is no such standard and we have to deal with the circle of confusion. This makes it all way more complicated. And although I agree we can investigate what the target curve should be by asking people what they think of it, there is no reason to include an obviously preferred bass bump in a target curve. Instead we should not ask people what they prefer, but what they think sounds more "natural", real life like.

This.

"Accuracy" (which is an ill-defined term, particularly when thrown around on ASR) is not synonymous with "preference."

 

[Yorkshire Mouth]

This.

"Accuracy" (which is an ill-defined term, particularly when thrown around on ASR) is not synonymous with "preference."

On this we agree.

The point that appears to be lost in some areas is that most of us here, for most of the time, for most products, couldn’t give a flying fig about what sound most people like. It’s almost the antithesis of what we come here for.

 

[Yorkshire Mouth]

If a clinical trial were employed to determine the efficacy, side effects, or magnitude of a therapeutic dose of a medication in a population, would it also be "unscientific" by these standards if they are based on subjective reports? If such a study determined that the ideal dosage was 50 milligrams per day (and factors like height/weight/age negligibly impacted therapeutic outcomes), would you really not consider this dose "scientifically correct"?

That's an interesting point, let's run with it.

Let's say we had a deadly virus, and we were trying to develop a vaccine. We run trials, and give 20,000 people the vaccine and 20,000 people a placebo. What next? How do we measure the efficacy of the vaccine?

I'll offer you two alternatives.

A - We measure the number of people in both groups who (i) test positive, (ii) require hospitalisation, (iii) require intensive care, and (iv) who die.

OR

B - We ask them how they're feeling.

 

[bluefuzz]

So you're saying that the preferred bass bump is accurate?

It’s as ‘accurate’ as we’re ever going to get. With all the vagaries of how headphones interact with peoples ears (no two ears are the same even on the same person) and with the added 'black box' of subjective experience a statistical subjective preference is the only viable approach.

I have set my tv to a white balance of 6500K, because that's the agreed standard,

You do realise that the CIELAB colour standard that your tv uses was essentially made in the same way? Basically a small group of people were shown coloured cards and asked about what they saw. This was done in the 1930s and updated in the 1970s. All colour standards from screen calibration to automobile paint is based on these ‘subjective’ tests.

Instead we should not ask people what they prefer, but what they think sounds more "natural", real life like.

I doubt that would change anything. It’s essentially the same thing.

 

[Yorkshire Mouth]

You do realise that the CIELAB colour standard that your tv uses was essentially made in the same way? Basically a small group of people were shown coloured cards and asked about what they saw. This was done in the 1930s and updated in the 1970s. All colour standards from screen calibration to automobile paint is based on these ‘subjective’ tests.

I'm afraid that's a tad wide of the mark.

The point is that any broadcasts or discs you see are mastered to that standard. So you know that, if your display is calibrated to that standard, you're seeing exactly the same image as the person who crated the final master for the disc/broadcast.

Now, if all audio mixing and mastering studios had speakers (& rooms) and headphones all EQd to the Harman Curve, you'd have a point. But they don't.

I've said this before. If we want a standard, a target, then surely it must be to hear from our headphones exactly (or as near as possible) what the engineer heard when creating the master. They mix it and say "Yes, this is how I want it to sound!" and the reproduction of that should be our goal.

Shouldn't it?

I mean, if you then want to EQ that to have more bass or whatever, that's your choice.

 

[preload]

That's an interesting point, let's run with it.

Let's say we had a deadly virus, and we were trying to develop a vaccine. We run trials, and give 20,000 people the vaccine and 20,000 people a placebo. What next? How do we measure the efficacy of the vaccine?

I'll offer you two alternatives.

A - We measure the number of people in both groups who (i) test positive, (ii) require hospitalisation, (iii) require intensive care, and (iv) who die.

OR

B - We ask them how they're feeling.

The answer is both A and B together.

For B, we would measure how patients are feeling using a structured questionnaire or other standard instrument to ascertain patient-reported signs and symptoms.

Are you STILL trying to argue that human reported data, like measured preference scores, are somehow not scientific? If not, what was your point here?

 

[Robbo99999]

@everyone involved in this current convo. In terms of the word "accuracy" and headphones it's hard to marry to the two due to the fact that we all hear headphones differently to each other due to anatomical differences (HRTF differences), so the word accuracy can only really come about if you manage to create your own personalised Target Curve based on your own HRTF, which is what The Impulcifier Project and also the Smyth Realiser attempt to do. Loosely, you could argue that Harman Curve is more accurate than other curves because it's based on an average head in an ideal listening room with ideal speakers (with a bit of preference study added to it), but in general terms it's really hard to use the word "accuracy" when it comes to headphones unless you've gone through The Impulcifier Project or Smyth Realiser. I hear reports that Smyth Realiser produces extraordinary results, but I've not tried either myself.

 

[Yorkshire Mouth]

The answer is both A and B together.

For B, we would measure how patients are feeling using a structured questionnaire or other standard instrument to ascertain patient-reported signs and symptoms.

Well, it's my job to arrange controlled Covid vaccination clinics in the UK for the NHS, for which I've had to plough through all of the literature on the initial trials, and decisions on whether or not to licence for use by the MHRA . And you're wrong.

Are you STILL trying to argue that human reported data, like measured preference scores, are somehow not scientific? If not, what was your point here?

Well, I've answered that already several times. If you can't be bothered to read the posts I've already contributed, I don't think one more will make a difference.

So no, tough. Go and read what I've said already.