Compensation Curve for InnerFidelity Measurements Dialog Part 1

New InnerFidelity headphone compensation curve.

First and foremost I want to thank each and every person who commented on my initial Compensation Curve for InnerFidelity Measurements post. An InnerFidelity target curve should be considered journalism and not quite science, but that doesn't mean I shouldn't be juried by peers. So I very much appreciate and take very seriously you comments about this curve. Thank you.

One thing I need to say right off the bat is that the graph tool will have the ability to select one of a number of compensation curves to apply, so you won't be stuck with the InnerFidelity curve I'm developing. Other curves will be: diffuse field; independant of direction; averaged raw; and some form of the Harman curve. Just wanted you to know you're not stuck with what I come up with.

I'm going to take the liberty of editing, reordering, and recapitulating the some of the comments and respond here. My apologies to those I haven't directly addressed...this topic is very complex and I'm just trying to stay focussed on the details at hand.

Tyll's Gone Full Arbitrary!

Submitted by -------------- on September 14, 2017 - 5:53pm
"Graphs are confusing, so I averaged them. The bass hump is controversial, so I got rid of it."

Submitted by Jazz Casual on September 15, 2017 - 11:59pm
I refer to Tyll's headphone measurements as an objective data point, but it appears that he's just arbitrarily gone and shifted the goal posts.

Submitted by castleofargh on September 15, 2017 - 4:59am
I don't get it. we start with a clear hypothesis about having the sound of speakers in a good room and go for it. then you don't like the results, which is fine, it's one idea with one room an one dummy head. then we enter the twilight zone.

Submitted by geniekid on September 15, 2017 - 7:35am
I kind of agree here. It feels like you just gave up trying to calculate the compensation curve at the low end because of an anomaly that couldn't be explained, although I suppose it doesn't matter as long as you're using the graphs for comparing different headphones.

From my point of view I've not been arbitrary at all...let me explain.

On getting rid of the bass bump
I think we can all agree that the bass bump is somewhat controversial. Sean Olive (named Tonmeister on posts here) commented at length about this. For the most part the size and shape of this bump comes from subjective preference and hasn't been studied enough for firm answers. Another source of this bump is that the small rooms in which most people listen to speakers get a bass boost as the wavelength approaches the room dimensions; the argument goes that because people are used to hearing it, they desire it.

The other problem is the bump my head measured does not have the same shape and amplitude as the boost measured by a calibrated measurement microphone. This may be due to room modes, or boundary gain of the head, or who knows what.

Lastly, there are numerous headphones, primarily planars, that theoretically should be flat in the bass and when I measure them they show essentially a flat response in the bass. Some run out of steam at very low frequencies, but it's clear to me that flat on headphones is simply flat in the bass. Here are some examples:

Audeze LCD3
170918_Blog_IFCompensation01_Graph_LCD3

HiFiMAN HE1000
170918_Blog_IFCompensation01_Graph_HE1000

Mr Speakers Ether
170918_Blog_IFCompensation01_Graph_Ether

Stax SR009
170918_Blog_IFCompensation01_Graph_SR009

Oppo PM2
170918_Blog_IFCompensation01_Graph_PM2

Koss ESP950
170918_Blog_IFCompensation01_Graph_ESP950

While there are all manner of rather arbitrary bass bumps I could insert in a compensation curve, it seems to me that flat is an objective standard and one which lends itself easily to viewer interpretation.

From ~300Hz up
In this area I don't feel I've been arbitrary at all. The curve in this area is basically the average of all curves measured. Except for some smoothing, I've not manually manipulated it all.

Still, it may not be right. Maybe it would be better to use only the stereo or left channel only, or maybe one angle is most representative of what should be heard on headphones. But which to pick? And wouldn't that be the more arbitrary choice?

My reasoning here is selecting any one measurement or narrow set of measurements will cause the resulting curve to have sharper features created mainly by the pinna reflections. While the pinnae of my head are very carefully designed to be close representation of the average human ear, they will be different from your ears. By spatially averaging all results I'm essentially removing any sharp features created by the pinna, leaving only a general approximation of what the ear hears in front of speakers. This resultes in a fairly smooth compensation curve.

A smooth curve is good because it will be less likely to accentuate or hide fine-grained features of the headphone under test. Many of the sharper features found in the speaker measurements change amplitude or center frequency depending on position of the head. If a compensation curve has a feature in it from one particular angle, it may cause a distortion in the curve if that feature slides to a slightly different frequency with a change in head position. A relatively smooth curve will be transparent to any sharp features in the headphone response regardless of where they occur.

None the less...

Submitted by castleofargh on September 15, 2017 - 4:59am
...now that I've lectured you like I knew anything, it's the best time to ask for something(I'm such a great negotiator). I'd like to see the raw front looking dummy(no angle), with stereo and with only one speaker as you seem to have both. to see how much the response changes on that head from correct "crossfeed"(IDK how that should be called)of both channels, against one channel for one ear like headphones.

Since you asked so nicely....

170918_Blog_IFCompensation01_Graph_CenterLeft

170918_Blog_IFCompensation01_Graph_CenterStereo

The problem I have with these curves—and the speaker measurements in general—is the big dip between 1kHz and 2kHz. After staring at the graphs for a long time it seems to me this trough is elevation sensitive and the feature most prominent when the head is straight ahead. It seems possible to me that the lack of this feature in headphones may be partially responsible for headphone imaging being high in the head. The hardest place to get localization is directly in front and this feature may show a cue that is sensitive in the forward direction. Unfortunately my data isn't really clear enough to make that call.

None the less, when we sit in front of a stereo system we want to hear the acoustic image fill the entire space between the speakers. If headphones were tuned for the center position, it may not portray information off-center properly.

500Hz to 2kHz
Submitted by mrspeakers on September 15, 2017 - 12:22pm
One thing I note is that the curve appears to create a notch/discontinuity around 1K. I suspect it very unlikely almost all these headphones have a midrange discontinuity like that, it's a pretty critical area, and so it looks to me like there's an issue there that might create concerns about midrange linearity.

Yes, as mentioned above I see a problem with any target with significant inflection in this area. In fact, I tend to think the Harman Target is probably right on 600Hz to 3.5kHz. For reference, both the Diffuse Field and Independant of Direction curves have deflections in this area as well.

170918_Blog_IFCompensation_Graph_IFDFIDCurves

I think we'll have to see how this shakes out over time as people get to use the graph tool. I encourage folks to listen carefully in this area and comment as time goes on.

Height of peak at 3.5kHz
Submitted by Tonmeister on September 14, 2017 - 11:00pm
3. The other source of controversy could be the trade-off between bass boost and 3kHz bump. Many open-back headphones cannot produce the preferred 4-6 dB bass bump that are easily achieved with closed back headphones. So manufacturers trade-off the the bass bump for a reduction of energy at 3 kHz. If you read the paper by Gaetan Lorho ( see http://www.aes.org/e-lib/browse.cfm?elib=14966) he took a headphone equalized to the diffuse field response (ie. flat bass) and had listeners adjust the level of the 12-15 dB peak at 3 kHz (measured in an ear simulator) according to taste. He found listeners on average preferred the 12-15 dB peak at 3 kHz adjusted down to about 3 dB if I recall. So listeners were essentially trading off bass for less upper midrange Several studies have shown that Lorho's target is less preferred to a target response like the Harman target that has more bass (4-6 dB below 125 Hz) with an otherwise DF-like response above 200 Hz.

Hm....interesting. Thanks, Sean. Personally I do tend to like this peak at about +12dB rather than +15dB...maybe you've identified why. At the moment I'm going to stick with the data gathered, but this is certainly something to think about moving forward. Eventually we might be able to have more than one target curve developed that covers the case of boosted bass or not.

3.5kHz to 8kHz
Submitted by brkitup on September 14, 2017 - 8:17pm
Thank you so much for your work on this. The graphs have traditionally not represented very well what the end user hears, and the new graphs are much more representative of how these headphones sound, especially in the treble. The one thing that still bothers me slightly, though, is the dip around 4-5kHz that seems common to all headphones when measured and compensated. I'm not sure how accurate that part of the compensation is.

Submitted by Argyris on September 14, 2017 - 8:55pm
Same here. It's smaller in width and depth in the IF compensation than on the DF or ID compensations, but it's still there. I wanted to see what this sounded like filled in, so I tried it on my HD 600 with a parametric EQ. I dialed in a simple peak centered around 5 kHz, with 0.3 octave width and 5 dB boost--it doesn't perfectly fill the hole, but it's a quick and dirty test of concept. It didn't sound horrible, like filling in the much larger and deeper trough left by the other compensations does, but it did add an uncomfortable edge and glare to everything.

Submitted by kais on September 15, 2017 - 3:00am
Currently the 5K range looks suspicious.

I quite agree, this area is tricky. It seems some headphone makers try to dial down this area to rid the cans of the piercing sound that can occur here, while others seem to keep the level up in order to improve speech intelligibility. I tend to see a lot of variation here. Fortunately, the way the numbers shook out kind of splits the difference and I do feel pretty good about it. Again, time will tell.

8kHz - 13kHz
The proposed curve does have a little detail in this area. This is likely some ear canal resonances. I was tempted to just erase them as they may create some artifacts, but as I looked at a lot of headphone plots I do see some seemingly common features in this area. Fortunately in the process of averaging all the speaker measurement curves these features are relatively small in the compensation plot so they shouldn't cause too much trouble and will likely reduce some of the noise in this region.

Finding Flat
Submitted by mrspeakers on September 15, 2017 - 12:22pm
In the absence of a standard "flat" compensation curve what about publishing a difference curve relative to your target? Musing out loud, people would be able to easily see where the response is on target or off and thus even if they're not able to intuitively process what the 4K dip really sounds would a differential plot would make it more intuitive?

Maybe I haven't been clear, but this is what I'm doing. The proposed curve will be subtracted from the averaged raw measurements to portray the difference between the two. The assumption is that if a headphone is tuned like speakers it will sound flat. Problem is, headphones aren't speakers, and there may be psychoacoustic reasons for headphones to have a somewhat different response at the eardrum than speakers.

Submitted by zobel on September 15, 2017 - 8:39pm
Nope. You cant measure what we see by putting a camera in an eyeball, any more than you can measure what we hear by putting a microphone in an ear.

Subjectivity rears its beautiful head here. The brain subjectively interprets the SPL data at the eardrum, and trough the entire inner ear as it is converted into a representation of SPL in nerve impulses. It wouldn't help to measure those impulses on the way to the brain either since the processor (brain) actually synthesizes our picture of the sound, with it's own correction curve, which is as individual as our set of ears are.

We will just have to make tests like Harmon did to find averages among listeners in comparing cans to a reference system.

Submitted by zobel on September 15, 2017 - 8:27pm
Mr Speakers asks the same old question that we all have been for many years, "Why not portray flat as a flat line, using raw data to provide the difference curve between measurements, and the perception of flat. It of course is not possible to do this scientifically or subjectively, but that attempt needs to be made to provide a more meaningful and representative curve. It would be an averaged curve, like Harmon's, based on listener's trials with a reference speaker system in a room, followed by headphone listening to determine the closest match to the room sound.

Yep. That sure would be nice. Unfortunately that task would be way out of my job description. I can only defend my methods here by saying I am unaware of any standardized curve for my head other than diffuse field, free field, and independant of direction, and those curves are simply wrong for headphones. Measuring my head in front of speakers, right or wrong, seems to me the best I can do. As Sean says:

Submitted by Tonmeister on September 14, 2017 - 11:00pm
In conclusion, I think the main reason for this bass bump controversy is elated to these 3 issues (no target will satisfy all tastes and all programs) and there is lack of published scientific evidence to support any one target. Hopefully, this will change soon.

Amen.

Odds and Ends
Submitted by kais on September 15, 2017 - 3:00am
I would like to see more headphone's curves compensated with this target, specially those models that are considered close to neutral like the STAX SR009. If you average a lot of those the result should be something close to a straight line.

I'm not sure I'll be able to average a big wad of headphones, but next week I will publish a couple dozen or so compensated plots so people can have a larger survey of headphones.

Submitted by markus on September 16, 2017 - 12:59am
I wonder how the frequency response would change if the dummy had arms and an upper body.

This is a good point. If you look at the proposed curve at the top of the page you can see the transition from flat to the upward rising portion at about 450Hz is rather abrupt. If my head had arms and a torso this might not be the case.

170918_Blog_IFCompensation01_Graph_HeadAndTorsoContributions

The dashed line labeled #2 in the plots above show the contribution of the torso to the ear drum response. You can see it does elevate response from below 200Hz. In the long run it may be that the IF compensation curve needs some messaging from 100-500Hz...but I don't want to be any more arbitrary than I have to be—I've already ruffled enough feathers.

Submitted by Argyris on September 14, 2017 - 6:38pm
The HD 600 and HP50 graphs match up pretty well with what I heard (and still hear with the HD 600) from both headphones, much better than the previous compensation curve.

Submitted by kais on September 16, 2017 - 3:28am
If the goal is that a perfect headphone shows a straight line in the measurement this new correction curve approach comes closer than the one we had before. Still some way to go.

Thanks. Seems to me it tends to warm things up a bit too much when applied, which means the proposed curve is a bit too cool. It might be interesting to play with tilting the curve some, but I'll wait 'til there's far more evidence to make that move.

Submitted by Pokemonn on September 14, 2017 - 6:24pm
Is web graph tool underdeveloping? Hallelujah!

Submitted by lerrens on September 14, 2017 - 11:52pm
Innerfidelity has the best headphone measurement database right now and the Graph Tool will make it super awesome! I'm so excited I have to create an account just to tell you this: Thanks for all your work! Really looking forward to this :)

Submitted by castleofargh on September 15, 2017 - 4:59am
...and congrats for the graph tool being made. that's really good news.

Hallelujah indeed. I'm stoked!

Please feel free to add more comments below, I will be watching carefully.

COMMENTS
donunus's picture

Thanks for doing this curve selection thing Tyll. Whether or not one target curve is more correct than others, it will still help us all in our thought process when analyzing which one mirrors what we hear to be more correct based on our own listening. I'm sure people will want a poll sometime to determine what the headphone listening crowd finds to be the most accurate compensation curve based on their own listening.

zobel's picture

Thanks for keeping the discussion going! I think your data is very valuable, and will be much more useful when we get to as close to a correction curve as possible, sans subjective testing, which would, when combined with your data, establish a theoretical flat 'curve' that would represent that elusive, generalized, averaged, subjective ideal, that a consensus would establish someday. Until then please carry on giving us repeatable hard data. Very worthwhile....Thanks.

steaxauce's picture

I'm curious how much perceived loudness and frequency response depend on the spatial information in the audio. In particular, I'm wondering if there really is a consistent frequency response mapping between speakers and headphones that will sound the same regardless of the recording, and thus, to what extent a target curve for headphones is really possible. Here's what I mean:

Say we have a function L(x, y) which tells us the loudness we perceive when we get an SPL of x at our left ear and y at our right ear.

With headphones, when a sound is played through the left channel, we get something like L(xh, 0), but with speakers, we get L(xs, ys), where xs > ys and the difference between xs and ys depends on frequency. How do you map the pair of SPLs (xs, ys) to an (xh, 0) that will sound equally loud, and how much does this mapping depend on frequency?

One question is whether the function is linear. That is, do we have L(x1, y1) + L(x2, y2) = L(x1 + x2, y1 + y2)? And, by extension (since the function is symmetric), in the case of speakers vs headphones, do we have L(xs, ys) = L(xs + ys, 0)? I have my doubts, and I wonder how well-behaved this function is and how much its behavior depends on frequency.*

None of this would be a problem if we only listened to mono content. But, I'm skeptical that developing a target curve to map the mono response of speakers to an ideal mono response for headphones will continue to sound ideal to the listener when we switch to stereo music recordings. The big question is, what implications does the behavior of this L function have for the ideal target curve?

The differences between the stereo images presented on headphones vs speakers depends a lot on the recording. Old, hard-panned recordings may call for a different headphone target curve than more modern recordings that tend to have less extreme channel separation. Even different instruments on the same recording may need to be compensated differently depending on how they're panned. Truly correcting headphones to get their frequency response to sound the same as speakers may require an "active" compensation filter to deal with any eccentricities that L function might have. Ironically, using a good crossfeed filter could make a lot of these issues less relevant.

Anyway, I'm just a lay person when it comes to audio and this is uninformed speculation on my part. Maybe someone has some insights or is aware of some literature on the topic? One question someone might be able to answer that might be relevant: When mastering music, do you typically apply any kind of EQ when switching from stereo to mono or otherwise changing the stereo image?

Tyll, thanks for making this effort. Very much looking forward to your graph building tool.

* Note: the math in this paragraph only makes sense in the linear domain, not the log domain (i.e. don't add decibels).

zobel's picture

Remember how the transparency of recordings was compared to a live microphone feed of the same performance? This compared (microphone- amp-speakers) directly to (microphone-recording-amp-speakers).

Live vs speakers is a much tougher trick. In just the right situation, listeners can be fooled by a recording on a playback system through loudspeakers, into thinking they are hearing actual instruments or voices. This has never happened with headphones, and by their nature, might be too large of a leap.

My point in this is: When we clamp transducers on, or in our ears, we give up too much of the natural environment involved in hearing to allow transparency in the process. No matter HOW accurately we reproduce frequencies, preserve transients and dynamics, eliminate every type of distortion, provide accurate spatial information....you know what? Yep!.. we know we have headphones on, and are hearing the sound through them. Give respect to our amazing brains for having at least this amount of discernment.

Which leaves us to 'the best we can expect from cans' being which sound most natural. Which reproduce a musical experience the best?
The musical experience of 'live performance vs live microphone feeds of it through headphones' would be the purest test. A transparent recording would also work of course.

Happily, we have recordings mixed and mastered on speakers that sound good over headphones, and vice-versa. Many, if not most recordings now use both in production, which helps ensure enjoyable results over cans. So I have these questions:

1) What is the ultimate goal of headphone listening...to develop as a parallel but necessarily less natural sounding sound system vs speakers?

2) Is this 'less natural sound system' superior in some ways, inferior in others? How so?

3) What factors are most important in development of this separate ....(and not equal).. sound system, in providing connection of the listener to the music via cans?
a)..widest and truest frequency response?
b)..lack of all types of distortion and coloration?
c)..dynamic and transient capability?
d)..focus and localization ability?
e)..comfort over long periods of time?
f)..portability? (efficiency, player compatibility) ?
g)..affordability?
e)..bling value, sex appeal :-) ?

I want it all, and I think the listening market does too. Frequency response is important, and as in speakers, small, narrow dips, or very gradual, smooth departures from perceived flat are preferable to peaks or a jagged response. Even narrow peaks become annoying over time, and are harder for the ear to accommodate. That might be due to the brain focusing on them, while being able to ignore what is not there.

Distortion, and that includes all those we measure, and the greater amount we don't, are in a bag of their own. A tough, complex issue here, with more factors to measure than we even know about, each having their individual particular affect on listenability...along with their interactive effects. Distortion is as critical, or more so.. than frequency response, assuming that SPL vs frequency is within the bounds of acceptability. Do we need IM distortion tests? What other distortions can we test?

The remaining factors will moderate, either supporting or negating frequency response and distortion achievements. What is most important c) through e)... to you?

Argyris's picture

I've long since abandoned the notion that headphones will ever recreate the sensation of a live performance. To me, the ultimate goal of any headphone is that it doesn't come across as annoying/fatiguing/painful to listen to for any length of time, and that it allows me to concentrate on the music without the headphone calling attention to itself. It's more about the absence of glaring faults (sometimes literally glaring) than it is about recreating a live performance, the feasibility of which is dubious at best when considering the multitrack recordings that comprise most recorded music.

Tonality is important to me, probably more so than any other quality, because I have pretty eclectic taste and I need my headphone to sound at least believable for any recording in any genre I choose at any given moment. Being able to cue up a Bach organ piece, some deadmau5, a movement from a Beethoven symphony, and Jethro Tull's Thick as a Brick right in a row without feeling like the headphone is much better suited to one selection over another is something that sets my best performers (my HD 600 and, to a lesser extent, my DT880) apart from the other headphones in my stable. It's the reason I keep reaching for these ones over the others whenever possible. Obviously colored headphones, tuned this way either intentionally or as a consequence of lax engineering, impede my listening enjoyment because they never disappear the way neutralish ones do--on some genres or individual selections the tuning issues might be less noticeable, but switch to a different recording and these issues jump right back out again.

Headphones to me are a necessary evil. Putting together a stereo system with equivalent tonal accuracy to my HD 600 would be prohibitively expensive, requiring tens of thousands of dollars in equipment and room treatment, assuming I had a suitable room available to begin with. And this doesn't even take into account the fact that I'm not at liberty to just blare music whenever I want. I put up with the weird spatial presentation of headphones not only because I'm used to it from decades of headphone listening, but because, despite its flaws, it works for me. Allowing me to enjoy my music is the problem I need my headphones to solve for me, and they do that, even if they're not (and will never be) perfect.

zobel's picture

Headphones are a thing unto themselves, and really... it is amazing they can sound so natural and so musical. I bought my pair of Senn HD600 from Tyll in Bozeman MT in 1997, there in his shop on Peach Street. They are still going strong after replacing pads a few times.

They hold up well for me in sound, but I have always missed the low bass, and transparent airiness, and efficiency, hence portability, that they lack. What is amazing for me, is their liquid and amazingly well textured midrange. Totally non-fatiguing, very comfortable, somewhat lacking is sparkle, but they are poor in the low bass. If the treble was that rolled off, they would have been dead in the water at the get-go.

I have a feeling Sennheiser has some magic they understand that goes beyond frequency response, and into the realm of distortion, that often makes them more transparent and musical, and less fatiguing than their competition.

Right now my all around favorite cans are Amperior. The voice coils are lighter, with much less impedance and inductance than higher impedance cans. They are quicker and easier to drive. Anything will power them well. Comfortable, portable, full bass extension, clean clear mids, and natural highs, and important to me, very dynamic and powerful when the music is..even with little players. Non-fatiguing with fair imaging, and I like the isolation too. I'm glad Tyll recommended them.

Stefraki's picture

First off, it was clear as day that the traditional curve needed to be abandoned. Like any headphone enthusiast, for years I have looked at a graph and 'mentally corrected it' – if you saw an actual flat graph you would think 'goodness, that's going to be bright' and if you saw one with attenuated higher frequencies, you would see it as 'right'. So that is plain silly, and did have to change.

But since no one's ear canals are the same, this is never going to be 'science'. We have to go with one persons impressions and frankly, is there someone more experienced as both an enthusiast and industry insider than Tyll? Why not just go with his take, and learn to 'mentally correct' this too? It's going to be less of a leap than the old curve regardless of the fine details.

But on the particulars – I do agree with leaving out the bass hump (for headphones anyway) as all people turning it up tells us is that the average person prefers elevated bass. I firmly agree that when you listen to an Audeze LCD model, the bass sounds flat to me, and it measures flat too.

On the 1k-2k thing, well my biases come in here. I have maintained for many years, having owned a few Stax Lambda models that have a little swell here, that a bit of extra emphasis here (on a graph) just subjectively makes the sound more realistic. The Harmon graph was something of a vindication of that subjective impression, but oh well.

Finally, IEMs need a different graph. Listen to an IEM that measures as having flat bass. Now listen to a headphone that measures as having flat bass. Notice anything? Unless my ears are radically different to the rest of the world's – the IEMs will sound incredibly bass light.

I don't know why this is, but I feel that comparative listening between any headphone and any iem that measures largely flat from 20–300 should be done. Tweak until the basically sound the same volume in the bass, and use that to create an IEM variant.

Unless it really is just me. I do have very large ear canals and have reflected that, due to the 'room' being such a small space with IEMs, ear canal size could have a really large effect on sound.

Anyway, really enjoy these sorts of articles, thanks Tyll.

bartzky's picture

As Sean and others have already pointed out, an anechoically flat speaker in a good room will produce a slope of approximately -1dB/oct. That's why a neutral bass should not be a flat line with a headphone, but should also form a light slope.

When plotting the smoothed measurement of the left speaker (L00) against the Harman curve to which the room has been equalized, a relatively similar picture emerges for the range below 200 Hz:

If you now subtract the Harman curve from the measurement and then drop the whole curve by 1dB/oct, you get the following picture:

I think something similar to the yellow curve represents a neutral bass better than a straight line.

If we now consider that in the case of loudspeakers we perceive each speaker with both ears, but in the case of headphones we perceive both channels strictly separately, the picture becomes more complicated. The following diagram shows the difference in frequency response between left ear (blue) and right ear (red) when only the left speaker is playing at 30° azimuth in a free field situation:

As you can see, the bass is almost unobstructedly perceived by the opposite ear. So in total there will nearly 6 dB bass missing without this crossfeed effect. It may be easy to add to this amount of bass to the curve now, but this might not be the right approach: this bass should arrive at the opposite ear and slightly delayed, but not just add to the ear near to the loudspeaker that is playing. The way to accomplish this would be DSP. For a compensation curve, it may be okay to neglect this effect. At least I don't have a better proposal at the moment.

To complicate matters: If you measure a headphone flat in the bass range, we might have a bass emphasis of about 3 dB in reality, due to leakage effects. Todd Welti's paper is particularly informative on this topic: http://www.aes.org/e-lib/browse.cfm?elib=17699
So the flat line might already be a hidden bass bump.

By the way: I sent you a binaural filter-setup based on your measurements of the Harman room a few weeks ago. I'd be very interested in your opinion.

wktenbrook's picture

Thank you, bartzky.

If you check out the earlier post...

http://www.innerfidelity.com/content/acoustic-basis-harman-listener-targ...

...and the followup:

https://www.innerfidelity.com/content/warren-tenbrooks-summary-head-meas...

You'll find my conclusions are very similar to your second figure.

The JBL M2 monitors used for Tyll's measurements have a programmable DSP that includes a frequency response/phase curve. I wrote Tyll and Dr. Olive that the JBL Professional calibration curve used for Tyll's measurements was a bit bass-heavy compared with another response curve RR1 (Reference Room 1) tested by Harman in their headphone research papers.

Your and my graphs are more consistent with the RR1 curve. Particularly see Figures 1 - 3 in second post and compare the delta between -1 dB/octave, curve RR1, and curve "twelti SP Preferred 2knob 6/1/16" used for Tyll's tests.

I wrote:

"Curve RR1 (violet) and 1 dB/octave slope (grey) again follow each other closely..."

I understand the controversy Tyll faces. Now that the JBL Pro M2 curve in the Reference Room is baked into his head measurements, it's difficult to back it out without appearing arbitrary.

bartzky's picture

Thank you for your reply, Warren.

I somehow missed your second article previously. It was a nice read!

I still think this graph taken out of your first article represents one of the best target curves to date:

Coincidentally the target curve (HFX) that we use internally for our headphone evaluation is astonishingly similar to yours (WTBT):

I derived the HFX from diffuse and free field curves mixed by directivity and a 1dB/oct slope.

I understand the controversy Tyll faces. Now that the JBL Pro M2 curve in the Reference Room is baked into his head measurements, it's difficult to back it out without appearing arbitrary.

In my opinion the arbitrariness startet by selecting the JBL Pro M2 curve. Getting rid of that and replacing it by a 1dB/oct slope seems eligible to me. I think you would agree with me.
I personally would even prefer to hold to the 1dB/oct slope instead of the RR1. They're pretty similar, but I think the 1dB/oct slope is more suitable to break the circle of confusion because it's such a simple curve.

Phoniac's picture

> So in total there will nearly 6 dB bass missing without this crossfeed effect.

If you add one speaker to another you gain 3dB SPL, not 6 dB. This is not simply double voltage, it is about acoustic power.

bartzky's picture

Thank you, Phoniac. Of course you are right! Sadly I can't edit my first comment.

kais's picture

because that's the only thing a compensation curve is about.
I'm only talking about frequency response in the following:

Listening to headphones is a multiple transition.
Listening to a natural sound source (e.g. a singer, instrument, anything) we usually look at and listen to it straight ahead, on axis.
This is about the same way the recording microphone is placed most times.
Now this recording his played back from stereo speakers angled about +/– 30 degree left and right from center axis, making the sound a bit sharp and unnatural in this stage. Recording engineers deal with this and try to compensate, with more or less success.
Then comes headphones, trying to emulate this sound.
Most headphones generate a sound coming from almost 90 degree (= from the side).
All this different angles completely change the frequency response, 1kHz and up, due to extremely different HRTFs.
Headphones making companies deal with this and try to compensate, with more or less success.
Finally headphone users deal with it by using a different playback level than the original sound source had.
Most dial up to a higher level, which increases the reception of mainly all lower and partly very high frequencies.

All this does not render frequency response measurements and compensation curves useless, but gives a certain amount of relativity, leaving room for personal decisions what a good compensation curve has to look like.
This is what every headphones manufacturer does when creating their house sound.
If we now had a compensation that renders an average of a lot of high-quality, commonly known neutral headphones as an almost straight line this curve would be something like a common sense between manufacturers and users. It could be used as a starting point for fine-tuning Tyll newly proposed curve.

wktenbrook's picture

All this different angles completely change the frequency response, 1kHz and up, due to extremely different HRTFs.

I expected that directional and HRTF characteristics would skew the response depending on center mono, L/R only, stereo, or multichannel measurements. However, Sean Olive reported they measured these configurations using the standard compliment of LSR6332 monitors in the Reference Room and there was very little change. Not sure if there's a paper on this. Certainly any effect will be more pronounced with increasing frequency.

markus's picture

I noticed that the 5 kHz area of the HD580/600/650 show a 5 dB dip in the 5 kHz area in the new compensation curve. I think it must be on the same level as 1 kHz, just like the flat coupler measurments on sbaf show it.

Jazz Casual's picture

I have always regarded your headphone measurements (along with everyone else's) as indicative. And though you might not have gone "full arbitrary", your explanation for not including the "somewhat controversial bass hump" in your new headphone target curve and opting for a flat bass response instead, looks very much like a subjective judgement call to me. But this is more journalism than science as you say.

I'll be referring to Innerfidelity's graph tool when it becomes available, and I think that having the facility to compare the different headphone compensation curves that it comes loaded with should be as interesting as it is inconclusive. ;)

Argyris's picture

As I mentioned back when the original findings were published over a year ago, I think a good subject for the new curve would be the MDR-V6/7506. It's an example of a headphone that draws a nearly flat line on graphs compensated with DF-like curves and yet which enthusiasts seldom hold up as an example of tonal accuracy. I've heard it on numerous occasions, and every time I found it glaring, shouty and sibilant all at once.

Not to mention, the treble drops off a cliff after 10 kHz--this might be useful for calibrating the 10 kHz+ range of the graph, since it's a clear example of poor treble extension.

Amusingly, one of my friends who has a 7506 listened to my HD 600 and said it sounded "tinny". I had to think for a second before I understood how he could have possibly come to that conclusion--what he was picking up on was all the upper treble that he wasn't used to hearing. From his perspective, it would be like turning up the 16 kHz slider on a standard 10-band graphic EQ all the way up.

castleofargh's picture

thanks you for all that.
I'm really only ever thinking about following through one hypothesis I find most relevant, using speakers as reference.
because we like them, and also to avoid making that damn circle of confusion even larger.
most of my all time favorite music was made on speakers for speaker users, so it seems like the obvious reference and target.
of course just saying that doesn't give a clear answer, if only. the bass for example is a big puzzle as we know. lack of bodily perceived low end, low end relatively to the rest of the FR, how much roll off and how soon was on the speakers used for mixing and mastering? did they push more by taste, because they expected the music to be heard on crap gear? or to follow some cliché about a music genre?...
I don't know what should be, if there is one value for all headphones/IEMs, all people, or if what I prefer is effectively flat like I believe it is for speakers.

so rich of all the not knowing anything, I hang on to what I have instead of piling yet another assumption over the rest. and that is the speaker reference. it's probably not better or worst than anything else, I'm not on some path of righteous neutral here. only that I don't know a all lot. I know that you measured a bump. that Harman believes speakers with that signature is the right choice. that their tests on headphones suggest how most people want a bass boost for preferred signature(how exactly is another matter as they usually didn't give full parametric EQ to the listeners).
so aside from not making the TOTL expensive orthos look a little less flat(down with the establishment!!!^_^), I don't see a reason to simply get rid of a bass boost.
Goldenears has been using a slight boost in the low end and it didn't shock me. I wouldn't say they have it all right(at least not for me), but if anything that's one more source who decided to part with the old standards nobody likes, and ended up with a bass boost.

as for the rest, well again I stick with speakers as reference until the end. if we assume listening to speakers, the actual source would be at 30° on each side, there is no reason to average the angles. it works just fine on actual speakers that way.
now as we lack the crossfeed on headphones, I'm not sure about using the response for one side or both. it's a case of "we're doing it wrong, but wrong how?".
an alternative idea would be to go with David Griesinger's cool little theory and practical tool. the full experience here, https://www.youtube.com/watch?v=1VaKZr208Dc but it's basically making 2 equal loudness contours while looking straight at one speaker then using headphone or IEM. it has the obvious and pretty cool benefit of not having the center image up or down depending on our body and headphone anymore. but it's also a little controversial as it aligns with nothing else.

anyway in the end on your graph tool we'll have DF and RAW for those used to that. the Harman curve for some idea of preferred signature, I feel like you could afford to stick pretty close to what you measured at 30° no elevation, because if your head was a dude, it would be his calibrations for a headphones FR like speakers. and then why not make your own subjective curve. the Tyll Thinks It's Flat curve. which to me is a little bit warm. and maybe a Joker Thinks it's Flat curve, for those who like brighter signatures. ^_^
I don't see the point of compromising so much with that curve started with a clear objective idea. it will be incomplete one way or another, but it will be something clear, or just your curve that might not even be your preference or perceived neutral.

wow that's one long opinion!

flatbass's picture

The point is not to worry about "headphones...that theoretically should be flat in the bass".

Flat frequency response under some measurement condition is not an inherent quality to strive for. When the research was done in loudspeakers, a flat on-axis response turned out to be important only because it was the most preferred subjectively. Subjective preference is the goal, not a philosophy about flat measurements. The same research also showed that the steady state in-room response of a such a loudspeaker includes the 200Hz bass bump.

Taken over to the headroom realm, that steady state response shows up as roughly what looks like the Harman curve, with the bass bump. This forms a starting guess as to what to expect of a good headphone response. Further research confirmed that the bass bump is subjectively preferred and *sounds flat* (neutral) to the listener. That's what's important.

It's true that we still don't have enough research to know the optimal size of the bass bump or whether the same size bump is universally preferred, but we do know that having a bass bump is preferred (and *sounds more flat*) over not. If uncertainty about the exact optimum means you should "arbitrarily" leave it as a pretty line at 0db for viewer interpretation, that same argument could be applied to argue that you might as well not have a compensation curve at all and leave the entire response to viewer interpretation.

zobel's picture

Thanks for pointing out the difference between what we measure in a room, be it through calibrated mics, or a dummy head, and perceived flat by actual people making subjective choices. They are two entirely different things.

I consider the RAW data from Tyll's dummy head to represent what is heard accurately both graphically and subjectively from 1000 Hz down. Therefore, no compensation needs to be applied to the data to achieve a representative graph. All this talk about a bass bump has no relevance to anything subjective, only to how the measurements appear when graphed.

The naturalness of room gain in the bass is to be expected. Any live sound in the room has the same exact gain in those frequencies. Without that gain in bass, we subjectively judge bass to be weaker than it naturally would be. Anyone who has ever run a PA for outdoor concerts knows all about the need to push much more power in low frequencies when outdoors, in what is an essentially mainly anechoic environment. Recording studios aren't anechoic rooms for good reason. No one listens to music inside in anechoic rooms. Outside, we boost bass to compensate. When headphones measure flat in the bass, that line represents perceived flat, with room boost present, or built in, therefore that is the target response for measured and subjective flat, or natural sounding bass. If... and only if, we measure frequency response in an anechoic chamber without the natural room gain in the bass, would it be necessary to apply a correction to the curve there to achieve natural sound that is subjectively judged flat.

Argyris's picture

I interpret leaving the bass hump out as exactly what Tyll said: since there's no consensus here, it's easier to just let people keep interpreting this part of the graphs the way they're used to doing rather than force us to learn how an arbitrary compensation feature figures into the compensated result.

I suppose you could argue that this applies to other parts of the IF curve as well, but I feel like this compensation does a better job than previously used standards from 100 Hz through 10 kHz, and I assume Tyll does as well or else he wouldn't have released it. If this all means I no longer have to do the mental gymnastics of interpreting a downward slope after 1 kHz and a wide, deep notch centered at ~6 kHz as flat (as was necessary with the ID compensation), then despite the uncertainty in the bass I still consider this an improvement.

Plus, it's not like Tyll can't come back after the bass issue is further hashed out and offer an additional compensation option in the comparison tool that includes the bass hump--I'm assuming that the this tool is just going to apply the compensation to the raw data in real time, so what's one more parsing option?

zobel's picture

I see that graph as a true representation of subjective flat.

Argyris's picture

I'm not really sure. Part of the problem is I don't have access to any headphones with perfectly linear bass extension, so I would have to extrapolate based on adding a low bass shelf to try to compensate for the rolloff on all my headphones. The closest thing I have is my DT880, which has surprisingly good extension for an open dynamic of its age, but which nonetheless runs out of steam around about 50 Hz.

Based on what I do have, and on experiments I've conducted using parametric EQ, I would personally favor a mild bass boost of between 3 dB and 5 dB below around 150 Hz (in other words, the Harman curve bass). However, there are too many variables here--including personal preference and how accurately I've compensated for the rolloff on my headphones--to be sure how well this maps onto objective reality.

So at the moment, I'm still watching the discussion as it plays out. Ultimately, I expect that we'll end up with two camps: one that favors the bass boost argument, and the other that favors flat extension. I think the best solution would be to have two compensations for users to choose between: one with the bass boost built in, the other without.

kais's picture

The amount of perceived bass is very much depending on the music you are listening to.
If you listen to non-remastered oldies you probably want much much more bass boost then if you listen to contemporary mixes of modern music.

kais's picture

The amount of perceived bass is very much depending on the music you are listening to.
If you listen to non-remastered oldies you probably want much much more bass boost then if you listen to contemporary mixes of modern music.

kais's picture

The amount of perceived bass is very much depending on the music you are listening to.
If you listen to non-remastered oldies you probably want much much more bass boost then if you listen to contemporary mixes of modern music.

zobel's picture

again we are not talking about a curve that matches the Harmon curve but only THIS...a curve that shows as a flat line for headphones that approximate that. This is what we should refer to as thw subjective flat curve, which I think already exists with the raw data below 1 kHz.

kais's picture

The amount of perceived bass is very much depending on the music you are listening to.
If you listen to non-remastered oldies you probably want much much more bass boost then if you listening to contemporary mixes of modern music styles.
Therefore the selection of music for testing has a great influence on the result.
The bass amount in the latest mixes is still increasing, and now, as the loudness war is over and hard limiting is no longer necessary (limiting evens out boosts of frequencies) it's possible and done to still have more.
So the Harman bass boost will be obsolete over the time.

flatbass's picture

Leaving <200hz as a flat 0db line is worse than having no compensation curve at all. It suggests to the viewer, most of whom likely have not read this nuanced discussion, that flat 0db at <200hz is the preferred target just like for >200hz. Even if research is not finalized on what size bump is optimal, there are some good candidates, and 0db is not one of them.

If Tyll feels conflicted about <200hz, a compromise would be to show a compensated curve only for >200hz. That would be less misleading than a line at 0db.

wiinippongamer's picture

"If Tyll feels conflicted about <200hz, a compromise would be to show a compensated curve only for >200hz. That would be less misleading than a line at 0db."

That's exactly what he's doing. The flat line on the bass area is the uncompensated response of the headphone. I hope you don't mean showing only >200hz on the graph.

Maybe he could put a slightly different color on the FR line for <200hz, to indicate a "no-guarantee" area, but it's probably not worth the hassle.

Pages

X