Harman tweaks its headphone target response

Those of you who follow developments in headphone measurement will know that in recent years Harman, with Sean Olive leading the effort, has been busy investigating and defining the ideal headphone target response – in other words, the headphone frequency response, as measured at the ‘eardrum’ of an artificial ear, that results in a perceived tonal balance preferred by listeners when replaying conventional stereo music recordings. Well, I’ve news of some recent developments on that score.

Prior to Harman’s research there were two schools of thought regarding headphone target response. The first was that it should, not unreasonably, match that of a sound source located in the horizontal plane and 30 degrees to one side of straight ahead – ie where a conventionally positioned stereo loudspeaker would be. This became known as the free-field (FF) target response. Logical as the FF assumption might seem, it was successfully challenged in the 1980s by the work of Günter Theile at IRT (Institut für Rundfunktechnik – Institute for Broadcasting Technology) in Munich, Germany, which identified as preferable the diffuse-field (DF) response, as experienced in a diffuse sound field where the sound has no particular direction. Of the two, the DF response held sway until Harman began re-evaluating the issue from scratch.

When Harman generated its first target response for circumaural (over-ear) and supra-aural (on-ear) headphones in 2013 its most obvious departure from the DF response was a low frequencies, which were shelved up below about 200Hz, thereby mimicking the in-room response of loudspeakers in an acoustically well-behaved domestic space. Since then Harman has generated two further iterations of this response, in 2015 and 2017, in which the LF up-shelving and other features of the target response are similar but modified in detail. Also last year, Harman announced a separate target response for in-ear headphones (insert earphones), which incorporates even greater bass boost.

Now that I have data for the latest Harman circumaural/supra-aural target response, I thought I’d share with you not just the new response but an overlay of all four Harman target responses so that it’s clear how they have progressed. You can see this in the graph below. In the key, OE stands for over-ear and IE for in-ear, although Harman has taken to calling circumaurals ‘around-ear’ (AE), which is both a more accurate translation and a more accurate description than ‘over-ear.' So in Harman-speak, AE equals circumaural and OE (on-ear) is supra-aural. All four responses are normalised to 0dB at 1kHz.

At high frequencies the latest OE (Harman: AE/OE) target response still quite closely resembles the diffuse-field target response although, compared to the 2015 version, the response peak at around 3kHz has reduced in level and moved up in frequency a little to 3.5kHz. Below 1kHz the 2017 target response has slightly increased in level, the 2017 and 2015 responses both being more markedly boosted below 200Hz than the original, 2013 target. The in-ear target response is very similar between 300Hz and 4kHz but notably features even greater bass boost.

Sean Olive explained the evolution of the circumaural/supra-aural target response to me as follows: “The AE-OE target curve evolved over the past four to five years. The first iteration was based on measurements of an accurate loudspeaker calibrated in our reference listening room. It was slightly modified with some method of adjustment (MOA) experiments where listeners modified the bass and treble, and also feedback from various tests with over 500 listeners from USA, Canada, Germany and China.

“The MOA tests led to about a 2dB increase in bass in 2015. In 2017-18, we reduced the treble around 3kHz by about 2dB largely based on feedback from listening tests conducted in Germany, China and USA. This new target curve was validated and tested against 31 different models, and reported in our recent AES Milan paper.

“The MOA tests have told us that the exact amount of preferred bass/treble will vary depending on age, listening experience, gender, and program material. Younger listeners and less experienced listeners generally preferred slightly more bass and treble than older, more experienced listeners. For senior listeners (55+ years old) we found they preferred on average even less bass but more treble than younger listeners. We believe this is related to hearing loss. This is an educated guess because we didn’t measure the hearing of the subjects (except Harman trained ones) and we need to test a larger sample.

“So the final target satisfies the majority of listeners, although it won’t be perfect for all programs and all people. We believe some degree of personalization is necessary to satisfy all listeners.”

When I begin my headphone testing for InnerFidelity, the latest Harman AE-OE and IE target responses will be used, as appropriate, to create corrected frequency responses, together with the FF and DF targets of old. This may seem overly complex but it allows individual readers to decide, on the basis of their own listening to reviewed models, to decide for themselves which target response is most appropriate for them. For example, I don’t generally like the full Harman AE-OE bass boost and find the diffuse-field correction more closely matches my perception of a neutral tonal balance.

COMMENTS
Gooberslot's picture

I wish you had added the DF response to the graph. It would have been interesting to compare.

KaiS's picture

The main problem I have with the Harman target, whatever iteration is referred to, it does not use an exactly defined sound field. In fact it seems you have to go to this worldwide unique single existing room for any calibration.
I wouldn't exactly call this a scientific approach, and the changes of the target over time show that it's more a trial and error thing.

Human beings can separate reception of direct and reflected sounds, measurement microphones, including dummy heads, cannot. The way diffuse or reflected sounds are integrated into the human reception of sound is not yet completely understood and can by no means be numerically specified.
That's the reason why diffuse field correction does not perfectly do the job. The approach is wrong or incomplete.

BTW: usual free field targets (or factory supplied correction curves) are often not referenced to speakers from 30° aside, but from the front. This translates even less to a "headphones sound like speakers" experience.

The theoretically only correct way to do this "speaker to headphones translation" would be binaural simulation of speakers in the room using each individual's HRTF. If you do this the targets doesn't matter, you just have to use the same target throughout the process.
Practically, and it has been done, it's not convincing because of many reasons.
Some of he more important ones are:
You have to add artificial signals ( reverb, room sound) to the music which is undesirable,
You do you need multi-axis head tracking because the human auditorial system relies on small head movements to locate a sound, if you skip this part the simulation will not sound realistic.
You need to exactly measure a multi directional HRTF of each individual person to do this.

My conclusion is: listening on headphones is a special experience (I do not even want to simulate speakers) and to make it enjoyable you cannot skip trial and error selecting your personal best sounding headphones.
If all headphones would sound the same (will never happen) you would not have that option.
Personally I have a comprehensive choice of HQ headphones which sound much more different then I would accept with speakers. Depending on the music I listen and my personal current taste I can enjoy all of them.

Tonmeister's picture

If you really want to replicate the sound field you can read the following papers and build the room. The acoustic properties of the loudspeaker and room are well documented and easy to replicate. I can even send you blue-prints.

1. http://www.aes.org/e-lib/browse.cfm?elib=14873

2. http://seanolive.blogspot.com/search?q=Reference+Room

Why did we make modifications of this target through "trial and error"? For several reasons:

1. The psychoacoustics of listening to stereo music through headphones versus loudspeakers in a semi-reflective room are different. There is no way you can arrive at a headphone target response simply based on a one-to-one measurement of a loudspeaker in a room. It is a good starting point but it will not sound ideal for reasons I think you already know. . Perception of bass is one example already addressed above. This is why we had to do significant tweaking of the target curve based on controlled listening tests.

2. Circle-of-Confusion Issues: Until the recording industry has meaningful standards that define the performance and calibration of loudspeakers, the program material is a huge nuisance variable. For this reason, a single target curve based on a measurement of a playback system will not satisfy all programs.

3. In_ear vs. AE-OE Target Curves: It became clear that the same target curve for AE-OE headphones does not sound correct for IE headphones.. When you plug your ear canal (occluded ear) with an IE headphone the acoustics and perception of sound are fundamentally different. It turns out listeners prefer about 4 dB more bass with occluded ear versus non-occluded. The reasons for this are not exactly understood but some possible candidates are occlusion-effect, and increased physiological noise.

I agree with you that the Harman Target Curve cannot satisfy everyone but our validation studies indicate it is generally preferred to the 31 other headphones we've tested for most listeners. Beyond that some form of personalization is desirable.

twelti's picture

I think you might be over-thinking the Harman target curve. It is not claimed to exactly reproduce some mythical perfect reference sound field. It is simply a curve which will generally be preferred by most people if used (with repeatable and standardized measurement method) as a target for headphone design. I would point out that the curves shown are exactly what was used in the listening tests. If used as a target for design purposes, we would normally use a smoothed version. Smoothed at 1/3 or 1/2 octave. Anything above 10 kHz will have some uncertainty due to measurement difficulties, and difficulties with the virtual headphone technique. Fortunately the most important frequency range for subjective preference is below this.

KaiS's picture

First I want to say that I appreciate your work and it's very interesting that someone researches in that field, decades after Guenther Theile introduced the controversial diffuse field headphones equalization.

A dummyhead's HRTF is extremely directional by nature.
The soundfield in your reference room is not directionally uniform too, e.g. reflections coming from the back or the sides have different frequency responses then signals from the front. If you now use the dummyhead microphone for measurements in your room the result is unpredictable.
Examples of clearly defined, uniform soundfields are freefield, diffusefield and pressurefield. These are uniform, controllable and reproducible. A "living room" with a stereo speaker setup is not.

Here is where some questions arise, how was your reference room calibrated:
What measurement microphone was used?
How was it placed and angled?
What signals and measurement arrangement did you use for calibration, sweep/stepped/wobbled sine, correlated or uncorrelated pink noise, TDS chirp, or ...?
Did you calibrate for each speaker individually?
What was the rooms target response?
Else...?
Did you use the same setup for the dummyhead's measurements?
How did you deal with the fact that you have two speakers and two (dummyhead) microphones?
I did not find those mentioned.

You partly work around irregularities by averaging several measurement positions and applying 1/3 or 1/2 Oktave smoothing to the result.
But this limits the usable resolution of measurements done with the system.

And here is where my concerns start:
you set new standards for headphone measurements, but I did until now not find mentioned the limitations of this new model (thank you so far for your statements above).
People start slavishly judging headphones based on the Harman target related measurements, taking it as absolute measure.
Some even use it completely wrong by just applying the curve on any system they have and things like that.
Considering the wide spread of your work I think you owe much more "education" and clear statements to the public, outside AES papers.
E.g. Innerfidelity should give some room for this.
Most people cannot go beyond "cookbook" knowledge. So there should be a clear list of dos and don'ts from you to avoid a widespread misinterpretation.

Finally: I am curiously waiting for correctly done measurements on known headphones, so practical results will show how far these correlate with my (and other's) experience.
Did you make some measurements public?

twelti's picture

Thanks KaiS for your comments and questions. Have you read "Listener Preference for Different Headphone Target Response Curves" AES 134, Rome Italy 2013. This has some of the details you are asking about. May I suggest you read that one and then pose any questions still unanswered? I may have to dig a bit to find some answers, as it has been 5 year or so, but I'd be happy to discuss.

Jim Tavegia's picture

I am having my hearing tested on the 16th as I know I have severe tinnitus and much hearing loss over 5khz. These are the results of 6 years of being in the Army and weapons ranges and being 71 in August. There is no one pair of headphones that will fit everyone and their preferences. Auditioning is critical.

My go to cans for my recording studio are my 2 pair of AKG K-271 (sealed) which for most people will seem bright, just not to me. My listening cans are my AKG K701s. I also own a pair of AT 50X and 40X, but due to my reduced HF hearing sound too bass heavy to me, but for the money I think most people with good hearing would like them. I gave our church my two pair of Sony 7506's as they are just not SOTA any more. I use a pair of Sennheiser HD-280's om my Yamaha digital piano which are fine for $99. I also like my very old Grado 80's which I have always recommended for my older friends based on price and being open-backed. You have to add the larger earpads to make them comfortable.

I will add two more pair of AKG K-271s for performers who come to my home studio to do some recording. If they find them too bright they can use the AT's I have.

There is no way I would spend over $200 on a set of cans without an audition first, meeting a target curve or not. Our hearing accuracy is a bigger factor.

Phoniac's picture

No mass-produced can will ever compensate your personal hearing curve (Army, 71...). Why didn't you go the most obvious and much easier route: EQ?

Johan B's picture

How can the IE curve have such a sharp fall off (Twice)? How can this be influenced without dampening? Over 15Khz nothing matters?

zobel's picture

It seems to me that the curve produced and shown here is fairly representative of what an averaged listening group, without too much hearing loss will consider the best overall presentation of SPL/Frequency, and probably for a good average listening level......When measuring with the same rig, of course.

The next step; chart the difference curve between the measured performance of headphones,(with this rig),and the best updated curve from Harmon. That would ideally give a flat line, if the measurement matches the generalized target curve. After all, isn't that what we want to see?

Tonmeister's picture

Thanks for the appreciation! You are correct: after a raw measurement has been corrected with the Harman Target Correction, a flatter curve is generally related to more preferred sound quality based on a large sample (n = 130) of trained and untrained listeners.. Below is a link to a slide from our latest AES Milan paper showing the relationship between deviations in dB from the Harman Target affect its preference rating.

The 31 headphones tested were categorized into four groups (Excellent, Good, Fair, Poor) based on their preference rating in a controlled listening test. In each category we show the average curve of those headphones (Blue) versus the Harman Target (green) and the error response curve (Red). The more the error response curve deviates from a flat line, the less it was generally preferred. The frequency range in which the deviations occur and the slope (dotted line) of the error curve also gives you a clue as to how the headphone sounds (ie. bright, dull, boomy, etc)

https://pbs.twimg.com/media/DeYs7HQW0AEYGkS.jpg:large

Cheers
Sean

sszorin's picture

I do not understand this - you gave those people doing the test headphones which have substandard, even grossly substandard, presentation of sound and you expected them to evaluate the quality of the audio signal, approximating Harman curve shape, fed into those headphones ?
If particular headphones have steeply rolled off treble or one note bass or if they have a volume hole in the middle frequencies then they are utterly useless as a tool for evaluation of an audio signal.
And another thing, what were the credentials of people doing this audio test ? I would say 95% of people do not care at all about audio fidelity / high fidelity, most of them even do not know what these words mean.

zobel's picture

Do you think that since all the excellent and good rated headphones in your research show less bass than the Harmon curve would indicate as best, could it be possible that the Harmon curve seeks more low bass response than what would be considered flat, or at least, preferred?

Tonmeister's picture

The way I interpret this is that the amount of bass below 50 Hz is not as critical as how accurate it is above that frequency. The critical area seems to be between 100-500 Hz where many headphones in the fair and poor sound category deviated further from the target than the headphones in the excellent and good category.....

zobel's picture

What headphones come the closest to a flat line as their difference curve to the Harman curve?

Max_Minimum's picture

I remember Tyll saying the Oppo PM3, NAD Viso HP50, and Focal Spirit (Classic and Professional) all pretty closely follow the Harman curve. I don't have the Focals, but I do own the other two and they sound very good to me, particularly the PM3. Unequalized, I like it better than even my LCD-X. I'm not saying it's outright superior to the LCD-X, just that I prefer listening to it when neither has eq applied. Oh, and it also depends upon not wearing glasses, or at least somehow getting them out from under the cushions. Otherwise the bass (on my head, at least) is ruined in both the PM3 and HP50.

DaveinSM's picture

I LOVE informative, interesting posts like this. Tyll’s original investigation and explanation of the Harman response curve was one of my favorite entries. This one is great and chock full of good information. Off to a great start and keep up the good work!

mariscosyketchup's picture

+8db under 100HZ? That's beyond low-fidelity, it's Fostex TH-900 bass boost level...that listener panel may be used to Best Buy boomboxes.
For speakers the B&K curve is way better than the Harman, for headphones, +2db bass boost under 150hz is perfect, more bass above that just clouds the mids if the mids are at the correct level.

Overall, I'm dissapointed because this curve promotes clouded mids and overly boosted bass.
And no, I'm not a deaf 80 year old person, I'm in my mid 20s and with healthy hearing.

briskly's picture

I do think this misses the point.
For speakers, saying that an ideal target exists is much like saying there is an absolute ideal loudspeaker directivity and room, and that all others are varieties of inaccurate. A more ambitious goal than any study I know of. Integrated over a sufficiently long time interval, the in-room response will be augmented by sound reflections the room feeds into the listening position. In Harman parlance, these are the first reflections and the sound power.

As for the bass shelf, the Q and frequency points were selected to reflect a slightly high subwoofer crossover point, as well as limit the midrange coloration. Q can be raised and frequency lowered to minimize this further. And of course, the result is only a mean of preferences with a handful of test tracks. A different set of test tracks could yield a less aggressively bass boosted preference. No one says that you have to like what someone else does either.

twelti's picture

The important thing is the shelving filter, that keeps the low end from getting boomy. Your opinions, such as "for headphones, +2db bass boost under 150hz is perfect" are your opinions, not universal truths. Our research is based on the preferences of MANY people, not one. Within that group, there are certainly different opinions, but we have to take an average to get a good target for the rest of us.

sszorin's picture

@ Low-fi bass boost : "+2db bass boost under 150hz is perfect, more bass above that just clouds the mids.."
Not exactly. That would depend on the quality and ability of headphones. If they have great resolution and dynamics/speed then bass frequencies stay in their place and do not flood into mid frequencies. I have Audio Technica W3000ANV which, according to FR graph, have about 5dB compensatory bass boost to deal with the perceived general headphones' bass deficit and the middle frequencies are not interfered with at all, they are exemplary clean. Because of this ability W3000ANV sound perfectly balanced.

Tonmeister's picture

As our studies have shown the amount of preferred bass below 100 Hz is highly dependent on program and the listeners' age, experience, and gender. Averaged across all listener the preferred bass for AE/OE headphones was 6.4 dB at 50 Hz. For a 45 year old experienced male the preferred bass was 4.3 dB which is closer to what is preferred for a loudspeaker in room.

For a 25 year old experienced male it's around 6.1 dB..

I already made the point that this variance in program and personal taste is a good argument for tone controls and personalization.

Since the bass boost begins below 150 Hz it doesn't promote or cause clouded mids. Most female and male voices have fundamentals that fall above this range and are not adversely affected.

Cheers
Sean Olive
Acoustic Research Fellow

Leffen's picture

"“The MOA tests have told us that the exact amount of preferred bass/treble will vary depending on age, listening experience, gender, and program material."

The classic story in this regard is when you're young, you boost bass and treble because it sounds more "hyper real", but as you age, you realize that that is artificial and getting in the way of the natural balances of the mixes. And you're better off just turning the music up, or doing nothing at all. Or making really subtle adjustments to improve the balance.

Ideally we want transparent reproduction with transducers, and then the ability to EQ or effect it as we please. Speakers should be tuned to be neutral for people with full hearing, and should be tuned to the preferences of wiser listeners who care about and listen for neutrality and "the tonality of real life".

For those with hearing loss, we can then give them personal corrective EQ curves and processing.

For teens who are having their early listening experiences and are lost in the dreamy beauty of it all, we can give them their own EQ's and processors if they want to make things sound more wild as they please.

That's the problem with those MOA tests.
I want an MOA test that only involves subjects who are mastering engineers or golden eared product designers or audiophiles.

I wish we had gotten Bob Katz's ideal headphone curve taken down before he left IF.

twelti's picture

We DO have some data showing the differences between young/old, and more/less listening experience. You can use that to modify the basic target curve.

sszorin's picture

Yes, what is happening with Bob Katz ? We need him.

Quizel's picture

Could you by chance make the new graphs with transparent backgrounds-instead of a opaque white or random color?

I'm guessing that's not a simple request as I see it no where. My hope is to one day be able to superimpose graphs of my choosing. That may need a completely custom user interface though... Perhaps it could be as simple as selecting the data sets one wants and hit the "draw all graphs" button. Providing the raw data of every recorded measurment could also work but that would most definitely be messy to look at for the average user -> even myself.

Hifihedgehog's picture

instead of allowing open dialogue, this new editor is sterilizing any comments from users who may have some advice and when he does comment, he chooses to be closed minded. Just see the article immediately after this one, where commenting has since been disabled and all comments (approximately 20) that do not meet his criteria have been deleted.

amadeogt's picture

...on that same post, because I actually enjoyed it (hadn't heard about TrueFi, and decided to try it because of the post). Unfortunately, comments were disabled.

Rafe Arnott's picture

If by "advice" you are referencing toxicity, then yes, that does not meet my, or anyone's criteria for discussion.

"Close-minded" I am not, aware that this site has been allowed to descend into mudslinging, yes, I am aware of that.

Comments moving forward will be moderated to maintain positivity and focus, not a place where people are concerned about being ridiculed for asking a simple question.

roscoeiii's picture

Can anyone point me to a good resource on why In Ear Curves differ so much from the other Harmon curves on both ends of the spectrum?

X