Big Sound 2015 Participant Report: Bob Katz

Day One: Blind Test and Headphone Shootout
I arrived at Tyll's on Saturday and was overwhelmed by the incredible assembly of gear that he assembled. Meeting the gentleman for the first time I know I have met a kindred spirit! Tyll has been dedicated to headphone technology officially for many years. While I'm way behind him on that score at least I can say that I have a collection of 25 or 30 cans and began my headphone adventures in 1972 with a set of Stax and a souped-up-by-me Stax headphone amplifier. So, let's say that Tyll and I went to different schools together!

Our first adventure was either a failure or a success depending on your opinion. It was comparing the best of delta-sigma DACs, the Antelope Audio Zodiac Platinum DAC, against a new version of the classic R2R technology, the unpronounceable Yggdrasil, aka the "Yggy" DAC from Schiit. My digital guru, Bruno Putseyz, has a number of disparaging things to say about the idea of resurrecting an R2R DAC, so I expected this comparison to be a piece of cake—I surmised the Yggy would sound colored and small. But this was not the case, and in fact, both Tyll and I failed the blind discrimination test, scoring no better than 50-50. We used a carefully-selected high-resolution reference recording from my collection that I know well, shortened it to about 20 seconds and looped it so we could compare one to the other blind in the fairest way. Still, sighted, Tyll and I both feel that perhaps the Yggy's got a little less depth than the Antelope. But maybe not, since we were unable to discriminate the two DACs blind and the Yggy is more than 4 times cheaper! I have an idea now for a listening test to proof the two DACs, but it's too late since I've already left Bozeman. If any of you readers gets an Yggy and has a good competing Delta-Sigma DAC, please contact me through Innerfidelity and I'll arrange to send you a special listening test signal that might reveal the sonic differences.

Headphone Shootout
For the headphone shootout, I concentrated on tonal accuracy first, including bass and treble extension, followed by attributes like spatiality, depth, soundstage, and lack of distortion (purity of tone). I used my own reference recordings which I know very well and actually have mastered for commercial release using a reference corrected loudspeaker system which is flat to a target response from below 20 Hz to above 20 kHz. The target has a high frequency rolloff and was created by me by listening to about 50 different reference recordings and ensuring that the average high end falls right in the middle. It falls right in line with my Revel speakers native high frequency response but has been smoothed to remove crossover and tweeter response anomalies in Acourate, which I'll discuss later in this review. So it's a very accurate loudspeaker reference; it sounds very natural.

My headphone reference at home is a set of equalized LCD-X and Stax 007-Mk2. They are equalized by ear to match the loudspeaker playback as closely as my human brain can do. But as you will see later in this review, there are probably scientific ways to do even better and so we are still learning how to equalize headphones. Here's a summary review: I didn't think anything I heard flat came up to the standards that I have set for equalized cans. I would prefer an EQ'ed version of every can I listened to—even the lofty Stax 007 MkII can be improved with a hair of EQ. That said, there were clear winners in the headphone contest when auditioned flat, and in principle, any headphone which sounds good flat is a candidate for minimalist EQ.

My number one choice: The Stax SR-007 headphones. I did not like the 009's at all. They weren't even contenders. They sound way too bright and thin on the KGSS amplifier. Perhaps with a sweet sounding tube amp the 009's would come into their best, but why spend almost twice as much to get an inferior sound? I think Stax went the wrong way with those 009 phones. Some listeners have written that the 007 sounds rolled off, but that was not the case with the KGSS amp. Perhaps it is the case with a tube Stax amp. The 007's had the best and most accurate tonal balance from bottom to top in Tyll's collection without EQ, but for sure they are lacking bottom without EQ.

Another reaction that surprised Tyll was the Spritzer-modified 007's which I brought along. We shot them out against the stock 007's and it was no contest: The bass on the modified 007 is more solid, extended, accurate. This mod is a no brainer necessity. The hardest part of the modification is opening up the cans and you have to be very careful but once you open them up it's a minute to put a tiny amount of blue tack or Mortite in the port and then reassemble them. Tyll agreed with me that this modification is a must do.

My number two choice: The Audeze LCD-3. I've now auditioned two samples of the LCD-3 and they are as different as can be. This sample sounds a lot more like my preferred LCD-X. I wager that unit-to-unit variations are greater than the differences between the two models. That makes it hard for someone to make a decision unheard and I advise you to go to a dealer that has both and audition two for-sale models if possible. This sample of the LCD-3 has the classic Audeze issue: missing a bit of the air frequencies, but like all of these great Audeze phones, it has a very pleasant tonal balance and is a great candidate for touchup EQ. The bass was as usual, Audeze-excellent. My number 1 and 2 choices are nearly tied, depending on whether you like high end more than bottom. Each one has its compromises and its strengths.

Number three snuck up on me. My first listen to the Enigma Acoustics Dharma was too quick. I dismissed it initially because after hearing the LCD-3's I got hit by accomodation and I felt the Dharma had a bit of an artificial hi-fi zing above 12 kHz, but on a second listen at Tyll's urging, after a bit of an ear rest, the Dharmas were very impressive from bottom to top with a little excess sibilance to tame circa 10 kHz. I definitely want to have a swing at these at home if I can get a loaner. Maybe with EQ the Dharma can beat the Audeze. It remains to be seen.

Number four is, surprisingly, the modified Sennheiser HD 800. To be honest, I think these phones are overpriced. When you open up a pair of these you wonder about the construction and the components. While a can like the Dharma have a specially made crossover and use separate custom-made low frequency and electrostatic drivers go for $400 less than the much-simpler Sennheisers! The modified HD800's, though, have a decent lower midrange, a reasonably tamed treble except for certain resonances excited by trumpets, for example, and a weak bottom that could easily be remedied with careful EQ, provided that your amplifier has the headroom. Bass frequencies require a lot more power than higher frequencies because of our ear's equal-loudness curve. People claim the unmodified HD800s have a lot of "resolution" but I think this is a result of a rising high frequency response. Not fair. If you like to "detail" at the expense of accurate tonality, then buy the HD800s and don't modify them.

Take a look at the video that Tyll shot to get my reactions to the rest of the phones, but the summary is that I didn't like the rest. It was either feast or famine folks, sorry.


Day Two: Amplifier Shootout
My new reference amplifier is my do-it-yourself build of the AMB M3 which I've written about in my column. I have been very interested in seeing how the M3 compares to the delectable commercial amps available for tasting at Big Sound so I brought it along in my carry-on. A suitcase loaded with a Prism DAC, a headphone amp and power supply and two premium phones. You should have seen the crazy time I had with TSA at the airport! In the time available I was not able to level match all the amps with Tyll's Fluke meter so most of these comparisons were matched by ear. Frankly, I was underwhelmed by the sound of the majority of the amps on display: Most of them sounded veiled, with a loss of definition usually in the treble but often in the bass as well. I'm a big fan of tight, responsive, low impedance voltage regulation which I believe translates to a speedy yet pure sound and I wonder if voltage regulation is the issue with some of these amps. My electronics guru, John Chester, says that when a manufacturer advertises he uses four stages of regulation for low noise, it's nonsense if you can't get it right with just two.

Perhaps these manufacturers aimed for use with some of the overbright headphones on the market, or to make "beautiful sounding" amps that complement some of the harsh-sounding loudness-war casualties in pop music. I can't blame them for trying. I love Coldplay's music but I can't bear to listen to their fatiguing recordings for long on my reference loudspeaker system. My friend, mastering engineer Bob Ludwig, who has made some of the finest-sounding recordings on earth, basically told me, "We can't tell Coldplay we would like to make a master that's lower in level than their mix!" So probably the source of the grunge goes back to their mixing, not the mastering. Bottom line: Some audiophiles want to listen to current pop and the only solution is a euphonic playback system. It's the only explanation that makes sense why a $4300 headphone amp sounds overly soft and fake-tubey, the Simaudio MOON Neo. It looks very impressive and has great features, but I'm afraid its sound does not come up to its looks. When Tyll listened during my comparison, he asked me, "are you sure you level matched them?" And I replied, "yes, to less than a tenth of a dB". And there's the rub: If an amplifier sounds "soft and warm", matching the voltage will not match the perceived loudness. Tyll and I agreed that we should manually turn up the Moon a few tenths in order to match the subjective loudness of the M3. Even then, on a level playing field, it did not come up to the sonics of the custom-built amplifier.

This situation repeated itself until I arrived at the beautiful red-faced HeadAmp GS-X Mk2. The photo of the Mk1 on Headamp's site does not do the Mk2 justice. It is now finished in a gorgeous "audiophile-grade" red finish as well as having sonic improvements. The GS-X has a pure tone, excellent transient clarity without any high end exaggeration, sweet midrange and good, solid bass. I felt that the GS-X Mk2 is in the same neck of the woods as my M3. I definitely wouldn't kick the GS-X out of bed, even though I felt the M3's bass definition a little bit better, we're picking nits here.

The only amp on display that I felt equaled the M3 in sonic quality was the custom-built KGSSSRE (Kevin Gilmore Solid State Special Reviewer's Edition E-Stat Amp), which you can't buy. I wonder if the original KGSS is still available. If it is, grab it, it's probably the best solid-state electrostatic headphone amp that's been made. Even though I could not audition the M3 and the Electrostatic amp with the same headphones, clearly all aspects of the two amps are in sync: transparency, impact, depth, purity of tone, soundstage... When you reach this league, the differences between Electrostatic and Dynamic melt away and you can just immerse yourself in the music.

Does this mean that you have to custom-build to get an accurate-sounding headphone amp? Not really. As I said, I really liked the GX-X Mk2. I haven't heard all the amps out there and I'm sure that some others are built with equal care and skill. Let's start to look for some more uncolored commercial models, ones with neutral, transparent sonics, since most people do not have the expertise to custom-build. Keep in mind that there are builders out there who will build an M3 for you. Builders are listed on the AMB website in the section on the M3 amplifier.

Day Two: Headphone Equalization: Seeking An Ideal Target Curve
I have began my EQ'ing adventures solely by ear, but at least I use reference recordings and compare them with the standard of a calibrated loudspeaker system and room: my own calibrated and flattened loudspeaker system that's been adjusted using a psychoacoustic measurement approach pioneered by Jim Johnston and interpreted by Dr. Uli Brueggemann. Brueggemann's loudspeaker/room measurement system is called "Acourate" and implements a variable measurement window according to psychoacoustic principles. Using 50 reference recordings that I know intimately, which I have recorded and/or mastered over the years, I have produced my own preferred loudspeaker target curve. Flat bass in the loudspeakers is my preference and a high frequency target with a gradual rolloff at the high end. It seems that many trained and professional listeners prefer this type of curve, while many untrained listeners like a greater bass boost and a bit less treble. This could be for entertainment (people love big bass), but keep in mind that I tend to listen to forte passages at an SPL around 83-85 dB, while many listeners use a lower level, so the equal loudness curves would dictate that listeners who play things softer need more bass boost.

But there's much more: Harman's curve contributes science to the art. If not for Tyll, his visit to the Harman facility, and his subsequent generous tutorial on the subject at Big Sound 2015, I would have been ignorant on this development. Since then I've read Sean Olive's seminal paper* and studied the research which has inspired Tyll and now me. Now I'm undertaking a systematic approach to quantify and evaluate these new curves from the perspective of a professional recording and mastering engineer. I hope that these observations take you all closer to your goal of headphone nirvana and I'll talk about them in upcoming episodes of Katz's Corner.

The principle of Olive's paper is that listeners prefer a similar response in their headphones to that produced by a flat-response loudspeaker in a listening room as measured at the eardrum position of a head and torso simulator. Olive's research also shows that untrained listeners prefer a curve with more bass boost than trained listeners and less treble than the loudspeakers measure themselves. In the next few months I hope to quantify and confirm those conclusions using my own reference recordings with my ears and visitors'.

Harman's listeners used recordings which they are not necessarily familiar with, while I have intimate experience as the creator of my reference recordings, which is a distinct advantage. The Harman testers liked a curve which sounds the most pleasing to them, while my goal will be to produce a curve which I think is most accurate because it most resembles the impression of the loudspeakers in my room and ultimately, natural, accurate sound. However, the closer a transducer is to the ears, the greater the perceived high frequencies and transients, and so it is logical that all of us will prefer a slightly more rolled off curve in the highs than what we hear from the loudspeakers.

Despite these variances, I think it will be possible to come up with a headphone target curve between two limits that you, the readers, will be able to take advantage of. We are living in interesting times, and I look forward to adventures with the Harman curve.

Our first attempt at EQ was very promising. At Big Sound, working with Tyll, I took my pair of LCD-X's and manually applied an EQ that is the difference between that of a measured LCD-X and the brighter of the Harman Curves, modified with Tyll's suggested 2.5 dB bass boost instead of the "typical" listener's 5 dB bass boost. Tyll was very gratified to find that we both think this curve sounds very good. I then began to modify the curve as auditioned on my own headphones. I found I had to brighten the 5 to 10 k range compared to the prediction. But that prediction was based on a different sample of LCD-X phones than my own. It was rewarding to discover after Tyll measured my personal phones, that they measures a lot more linear than the previous sample and so it was logical that I would need to push those frequencies. Clearly this procedure brings science into the art! In future episodes of Katz's corner I will start with Tyll's measured curve of my personal cans and algebraically subtract it from the Harman curve to derive a scientific EQ which I will begin with.

Another point is that bass and treble are yin and yang frequencies. If you boost bass you will perceive less treble and vice versa. This is why Quad invented the "tilt EQ". It's also how the smiley curve came into being! "More treble, Charlie. Now give me more bass. More treble. More bass." My brother's car inevitably had a graphic equalizer shaped like: A smile. But when you have boosted the bass, and suddenly think you need more treble, the first thing to do is not to fall into that trap, to reduce the bass a bit rather than add more treble. So it's going to take some time for me to evaluate this LCD-X EQ at home, decide whether the 2.5 dB bass boost is right and evaluate the treble in the context of the bass.

As of this writing, I haven't yet made a new EQ curve based on the exact scientific difference between my own LCD-X and the Harman target, but that is definitely in the near future. Here's a screenshot of the work in progress:


Bob's Harman-based LCD-X EQ. Work in Progress.

This eq resembles the shape that I had previously created, but the turnover points are quite different. Remember, this is a difference EQ, the amount of correction needed to be applied to a headphone's response to yield the Harman target curve. It is also a hand-tweaked curve because it was originally based on a different pair of LCD-X. To see the actual response of the headphones, we would need to display the resulting response of the headphones with this EQ. In future episodes of Katz's Corner I'll be sure to do this, using actual measurements.

Chinese Winners?
Just before I left Sunday, Tyll showed me a pair of inexpensive Chinese phones that I think cost about $69.00 and reside in his wall of fame. Tyll, what model was that? I want to get a pair. These are perhaps the best cans I've heard under $300! They'd go great with my iphone. My previous "inexpensive" winners are the Sennheiser HD280 Pro: I own three pairs! Perhaps these Chinese cans beat the $95.00 280s. What a wonderful visit and what a wonderful discovery.

Ed Note: Those were the Noontech Zoro II HD, Bob. On the Wall of Fame currently; great little headphone/headset.

* Audio Engineering Society Convention Paper 8994, "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses", by Sean Olive, Todd Welti and Elisabeth McMullin.

Equipment List

Front End
NAS - Synology DS414 ($479)
Renderer - Aurender W20 (~$17,600);
Digital Distribution Amps - Four ATI DMM100 Digital Matchmakers
and one DDA212-XLR digital audio distribution amp ($1450).

Power Conditioning
PS Audio, two P10 power regeneration station ($4999) and four DecTet conditioned plug strips ($499).

AURALiC Vega DAC ($3499)
Simaudio MOON Neo 430 HA ($4300 w/DAC).
HeadAmp GS-X Mk2 ($2800)
Schiit Ragnarok ($1699) and Yggdrasil ($2299)
Burson Audio Conductor Virtuoso ($1495 w/PCM1793; $1995 w/ESS1908)
Woo Audio WA-234 ($15,900)
Antelope Audio Zodiac Platinum DSD DAC, Voltikus Power Supply, and 10M Rubidium Atomic Clock. ($13,045)
Apex High Fi Audio (TTVJ) Teton ($5000)
Eddie Current Black Widow ($1248)
Violectric V281 ($2299)
Bakoon HPA-21 ($2995) current output headphone amplifier.
KGSSSRE (Kevin Gilmore Solid State Special Reviewer's Edition E-Stat Amp ($Unobtanium)

Sennheiser HD 800 ($1599)
Audeze LCD-3 ($1945) and LCD-X ($1699)
JPS Labs Abyss AB-1266 ($5495)
Stax SR-009 ($4450) and SR-007 ($2350)
HIFIMAN HE-1000 ($3000)
Mr. Speakers Ether ($1499)
Enigmacoustics Dharma (~$1200)
Audio Zenith PMx2 ($1398)

Digital cables by AudioQuest.
Cable complements for wiring entire systems will be from: Nordost; JPS Labs; WyWires; Cable Pro; AudioQuest, and Cardas.

Headphone stands by Klutz Designs

Ulrich's picture

Hi Bob, Tyll,

At Harman Advanced Technology Group, we found that measuring at the eardrum is actually very inaccurate above 1-2kHz. There are substantial deviations from person to person, and even more so between any person and standard ear simulators. These responses caused by the individual anatomy are compensated by the brain! How would we be able to have a sense for natural sound otherwise! I'd suggest to read my AES paper for more details related to headphone EQ (138. AES, Warsaw, paper 9274). I hope you get a chance to test the AKG N90Q soon!


ashutoshp's picture

Any possibility of making a cheaper version?

Bob Katz's picture

I just skimmed your paper, Ulrich. Good work. Well, it seems to me that the individual's perception is the ultimate way to make a judgment. However, Olive's paper makes some sense to me in that when we are missing the response of the torso, concha, etc. in the attempt to make the headphones sound like the loudspeaker. You are saying that the individual anatomy is compensated by the brain in loudspeaker reproduction but it seems to me that does not refute the Olive paper's conclusion that when the effect of the individual anatomy is not present, then the sound is wrong when reproducing recordings meant to sound good on loudspeakers on headphones.

I must admit that so far in my experiments this week, the Harman curve, even the "preferred listener" curve, appears to sound too bright. Either that is because of radical differences in my physiology from that of the HATS OR that Olive's method of defining a "flat loudspeaker" is very different from mine.

Both Olive and I are using Revel loudspeakers, though of a different model, and my equalized high end target nearly parallels the minus 0.5 dB position of my Revels, which definitely shows a measured rolloff. Consequently, if Olive's corrected loudspeakers truly measure flat, then they will sound much too bright. I know he is aware of that so I need to re-read his paper and get a clarification.

Similarly, you cite a loudspeaker compensation curve with rolled off treble in your paper that recognizes that there should be an HF rolloff. And perhaps also a bass boost. I'm only at the earliest stages of evaluating the Harman curve over here and tweaking it so I'll keep reporting on the progress or lack of same. I'm currently at the stage of adding a linear HF rolloff to see if it works while retaining the basic Harman shape.

Bob Katz's picture

I think I just found the fatal flaw in the harman response. It's figure 3 in Olive's Convention paper 8994. If these loudspeakers are truly equalized to an in-room response that's flat, then they will inevitably sound very bright and thin with any decent reference recordings among the 50 that I cite as standards. My measured target response is flat up to 1 kHz, 6 dB down at 20 kHz, linearly sloping down from 1 kHz, using a psychoacoustically-derived variable FFT window that is less than 10 ms wide at 20 kHz. This response has been endorsed by many practitioners and listeners using this measurement method.

Given that and the measured reported flat response of the Harman lab's speakers, I should start inserting a similar rolloff into my Harman curve-corrected cans and do some further listening. That's now on my agenda. Personally I have been very dissatisfied with the sound of any loudspeakers that measure flat. Inevitably they sound bright, thin and forward.

Bob Katz's picture

Harman's research is seminal and between Olive and Ulrich, I think we're going to end up with a new frontier in headphone design. Nevertheless, I've found what I believe to be a fatal flaw in listening and assumptions in the Harman report, leading to a target curve which is much too bright.

Figure 2 of the Harman paper shows the in-room amplitude response of a loudspeaker prior to equalization. These are loudspeakers designed and voiced by Harman's own Kevin Voecks, who is a reknowned audiophile designer. Kevin has spent years perfecting the art of speaker design both by measurement and by ear. I know, because I own a pair and respect them very much. Yet, they measure, as can be seen in figure 2, about 6 dB down at 20 kHz in-room response.

So it seems in principle that Voeck's intended response would sound very good without any alteration in principle. Turning them up 6 dB is a radical change, so only one response must be right. I vote for the original intentions of the loudspeaker designer, who after all is a Harman employee and these are now officially Harman-manufactured loudspeakers.

A much more logical approach to equalization would be to try to parallel the native high end response of this excellent-sounding loudspeaker, smooth some of the crossover variations and the low frequency room-based anomalies, and not fix what ain't broke!

Anyone who listens for a living would tell you that flattening these loudspeakers' high end will result in a sound that is too bright and too thin. So I vote for Kevin Voeck's initial frequency response shape. I suggest Olive retake his paper with that in mind.

Therefore as my next experiment I intend to impose a -6 dB at 20 kHz linear target on top of the Harman curve and see how that sounds. Stand by for the fireworks.

Again, I intend no offense: I find the reasoning and scientific approach of the Harman paper to be impeccable and hope that Olive will respect my reactions and consider an alteration to an otherwise-incredible advance on the state-of-the-art.

Ulrich's picture

My point is that ear simulators and artificial heads are insufficient tools to measure headphones at high frequencies, because of the individual differences, which are compensated by the brain. When you wear a headphone, the anatomy stays intact, except the torso (which is included in a pair of binaural filters in the N90). Also, I think you missed something in Sean's paper. He equalizes the speaker to flat initially, but then dials in a room target curve with a bass boost and HF rolloff (I am using 3-4dB rather than 6dB but it's made adjustable in N90).

Bob Katz's picture

Dear Ulrich:

a) I'm still skeptical, and still leaning in the direction of the Olive paper. How can we scientifically prove which approach is more accurate: Olive's paper or yours? Your claim is that the brain continues to automatically compensate for the anatomy when wearing headphones. In that case it seems a more flat-measuring headphone should be what's correct, no?

While I cannot academically decide right now between which approach seems to be correct at least I can make practical experiments using the Harman curve. So far this week my own personal manual EQ sounds better (more accurate) than the Harman curve but I'm still deep in the middle of experiments. Since I'm using a custom FIR filter I have to settle an issue with a minimum phase approach and where the impulse resides in my convolver to ensure I'm doing the filtering correctly. Then I'll get back to the experiment.

If you'll send me (at bobkatz[atsign] your personal email I'll keep you apprised of my reactions to the Harman EQ. I haven't written Sean directly and will wait until I have more data.

b) Thanks for making it clear to me what Harman was doing with loudspeaker compensation. Clearly I did not study Olive's paper sufficiently. Then I have to look elsewhere why the Harman curve is sounding too bright to me. Possibilities are that the three music samples they used were poorly chosen or the choice of ordinary listeners instead of trained listeners... or as you say, individual pinnae and concha responses vary widely as well as the high frequency responses using an artificial head.

Pity. A royal pity. I'll keep you posted.

ultrabike's picture

I don't think that Ulrich is saying that ones brain compensates for the anatomy for headphones only. I think it applies to speakers as well. In other words a perfectly flat response using a speaker and standard measurements will yield all kinds of responses if we bury the mic inside different head and ear models.

You could then say that a perfectly ideal speaker would yield different responses inside the ears of different individuals. The brain of each individual would have to compensate for these differences, and perhaps always does as our anatomy changes as we grow old.

The approach I see in speaker characterization is to remove anatomy and brain compensation, and concetrate in sound reproduction before it hits the body as this reflects how "real life" sounds will be processed by everyother individual. I would advocate a similar approach for headphones.

With headphones however, this may become increasingly difficult as the phones bypass different parts of the human anatomy (not to mention issues with sound absortion for measurements as making mini anechoic chambers seems a bit of a complex problem, though it can be approached). We further are not adding the issue of cross channels not present in the delivery of sound through headphones.

Regarding minimum phase. This means that the zeros of the headphone transfer function lie inside the unit circle. Which means that an ideal inverse digital filter with poles cancelling these zeros of this filter would be stable. I do not feel this is a good approach. Some zeros maybe too close to the unit circle and may move a little in frequency depending on how the phones rest on our heads. The results might be a bit catastrophic. One can try an approach that does not attempt to fully invert the response but merely apply small modifications that would improve headphone response. In such a case, we could even remove the requirement that the inverse filter need to have poles (IIR) and come up with an FIR approach which is much more stable as well. Given enough taps, a target curve, and a decent cost function one could probably come up with an automated approach that would do a half decent job.

Note also that a headphone may not be minimum phase, just approximately minimum phase.

Anyhow. Those are my current thoughts right or wrong.

Bob Katz's picture

Given those premises, I still don't see what's wrong with the Harman premise that the response as measured at the eardrum should mimic that of a set of loudspeakers. Even if each head, pinna and concha are different, it seems to me in theory that if we had the identical head for the loudspeaker and headphone measurements and the identical test microphone, that the ideal would be to match those two responses. If I'm getting the chicken and the egg mixed up, I'll bet I'm not the only one!

As for minimum phase, I believe that transducers are minimum phase and an equalizer that corrects the amplitude response will also automatically correct the phase. Of course I'm using an FIR approach, which can do minimum, linear or mixed phase without any problem. I simply have to deal with the fact that the impulse I have currently of my filter is at 0 seconds in a system designed to use an impulse in the middle of the impulse response. If I convert the impulse response from minimum phase to linear phase in order to move the impulse to the middle of the impulse response, will it screw up the minimum phase aspects of the filter? FFT is not my strong point!

ultrabike's picture

You could use a mic inside a head and measure speakers with the correct matched compensation. The issue then becomes that of positioning. As you move the mic off then the shape of the head would color the sound differently and you would require different compensation curves. This would be problematic depending on how one defines listening position or if one deviates from that, not to mention polar plots would be a bit more involved. In headphones, Tyll uses an average of different measurements since slight deltas in headphone repositioning result in variances in the measured response, particularly in the high mids and treble area.

For minimum phase, I think I factored out a sampled AKG701 response measured by Tyll some time ago, and found a few zeros slightly outside the unit circle if I remember correctly.

Any approach you take in correcting the magnitude response will do wonders for the phase response as those two are somewhat related.

Also, my intuition tells me that you should move the main tap of your filter outside 0 seconds. That would not necessarily result in linear phase behavior, but it will probably do better than a forced minimum phase filter. What will likely happen is that you may experiense a slight delay from input to output. But perhaps much better performance.

An ideal filter would work on previous ("post-ringing"), current, future ("pre-ringing") samples. A system cannot know the future, so future samples are out of the question (due to causality). What is done is to delay the main tap so to force "future" sampes appear at the "present"... Or something like that I think.

Bob Katz's picture

I just heard from Dr. Uli Brueggemann, inventor of Acourate and Acourate Convolver. He says that the Convolver can easily play minimum phase impulse responses whose impulse is at time 0 and the result will be correct. No need to convert and it is not a "forced minimum phase" filter.

Delaying the impulse could result in losing some of the information at the tail end of the fixed impulse response length as it will be truncated.

ultrabike's picture

I don't think the issue is that a minimum phase filter can't be played or can't correct for headphone issues. It can. But restricting the correction to be minimum phase is IMO unnecessary and in some cases, sub-optimal.

If you allow yourself to place zeros outside the unit circle in the correction filter you have more variables you can play with... if the SW tool you are using allows it.

As far loss of some information with regards to minimum phase vs. non-minimum phase, I think that in both cases you will loose information because these FIRs are infinite in length. In the case of a minimum phase filter they tend to go to positive infinity and in both cases you will need windowing, i.e. cut some information off (though usually that which is cut is a relatively small fraction of the total response)

However, I think you might need more samples w a minimum phase filter vs a non minimum phase one for the same performance in many cases.

In any case, I think if you are restricted to minimum phase filter correction you are likely to do fine job, and life is good.

Ulrich's picture

Ultrabike has explained it correctly. We (experienced listeners) are able to judge a neutral loudspeaker correctly, regardless of the individual response at the eardrum, which varies vastly from person to person. Which means we need to characterize a headphone by the sound it emits, not the sound that arrives at an (artificial?) eardrum. There is even more going on in our brain when we analyze the ear drum signal - a separate estimation of timbre and vertical direction of a sound source. Highly nonlinear processing! By the way - a headphone response is in most cases not minimum phase (depending on the enclosed volume in the cushion). A single frequency response is not sufficient to measure the headphone in these cases. You need a microphone array between driver and ear when wearing it. Quite similar as in loudspeaker measurements, where the on-axis curve does not always reveal the truth.

Bob Katz's picture

I think both Ulrich's and Olive's approach have to be considered as theories until proven. So far Olive's paper includes lots of experimental verification while Ulrich's is just a set of logical postulates.

Since Olive's approach is well documented I intend to put it to the test using subjective comparisons of the sound of equalized headphones versus my loudspeakers. It will be a very useful experimental approach.

To say that definitively the brain accommodates to a given experience can be taken to the logical conclusion that it doesn't matter what frequency response the headphones should have. Whereas when a pair of headphones sounds more natural and correct to me, my happy nerve lights up and I'm very selective about that.

The book has not been written yet and I urge all readers of this forum to keep an open mind as we sort through the theories and the practice.

ultrabike's picture

"To say that definitively the brain accommodates to a given experience can be taken to the logical conclusion that it doesn't matter what frequency response the headphones should have. Whereas when a pair of headphones sounds more natural and correct to me, my happy nerve lights up and I'm very selective about that."

Not necessarily. Our brains constantly have access to "real world" or "live" sounds. A train, someone's voice, a musical instrument, and so forth. These all serve as references. If a headphone's FR is messed up, then a violin through it would not match your experiense of how a violin is supposed to sound based on what you have heard real life. Changes on how you perceive a violin will happen, but will proly be dramatic only if you experience hearing loss or something like that... and even then, with time, and given real world reference, your brain will do it's very best to adapt. I think Tyll mentioned that in an article here at IF as well.

Bob Katz's picture

.... and affected by the physiology of the ear, it is an incomplete statement to say, "characterize it by the sound it emits." Ultimately you will have to apply an equalization which you can justify by any means you wish, but the headphone will have to be equalized. There is no flat transducer that you can use to measure the headphones that will make them sound correct. And so there has to be a rationale for the equalization.

I'm looking forward to seeing the fireworks between Ulrich's and Olive's efforts. As for now I'm going to see how well Olive's approach works in a practical situation. Anyway, I have no other way to go since I'm not privy to Ulrich's efforts, which are also subject to a patent application.

detlev24's picture

"[...]Which means we need to characterize a headphone by the sound it emits, not the sound that arrives at an (artificial?) eardrum.[...]"

A very interesting point, I think. Did you already approach this type of measurement and if so, what is the biggest challenge so far?

audiofly5's picture

''Ultrabike has explained it correctly. We (experienced listeners) are able to judge a neutral loudspeaker correctly, regardless of the individual response at the eardrum, which varies vastly from person to person.
Isn't that correct that we can also distinguish a good speaker regardless of the acoustic environment? In other words our mind tries to figure out what the source sounds like ( example: a bird singing in the woods) and 'filters' the environment out, in order to be more accurate at detecting what it is.

Bob Katz's picture

The more I learn the more I learn. In a later response in this thread, I will learn that the ACTUAL Harman curve is much closer to what I was concluding was needed, so Olive and I are very much on the same page.

But (caveat) I have not yet performed the longterm eq'ing experiments and my conclusions are preliminary. It will all come out in the wash. Thanks for your patience.

Bob Katz's picture

My conclusions about Olive's paper were not complete. Essentially I was not comprehending the whole story about this headphone target curve! Tyll, I suggest you reread the Harman paper, Convention Paper 8994. Notice that it all leads to paragraph 5.3, which demonstrates that the Harman curve in Figure 17 is really an INTERMEDIATE curve and an additional EQ on top of that must be applied to yield the preferred listener response.

This is exactly what my investigations this week are starting to show. In other words, my initial conclusions that the "Harman curve is bright" was based solely on Figure 17. Not fully reading the conclusions of the paper. Paragraph 5.3 makes that clear. So I was going in the right direction even without having fully absorbed the paper and Olive and I are really on the same page.

That's at least part of what Ulrich is trying to tell us.

1) Olive first measured the native in-room response of the loudspeakers, which shows an HF rolloff of about 6 dB at 20K.
2) He then equalized those loudspeakers to measure flat. At that point he measured the flat loudspeakers through the dummy head to yield Figure 17.
3) Then Olive evaluated listeners various responses to the music under test with these flattened response loudspeakers, and allowed the listeners to adjust loudspeaker bass and treble and headphone bass and treble.
4) Next he concluded that the "RR1 response" of his loudspeakers very closely resembles the desired result. Basically, RR1 is a loudspeaker with somewhat measured rolled-off treble. Which I agree with! It is pretty much the native response of the Revels. I disagree with the bass boost but I'm very good with the overall shape. In my room I engineer the rolloff turnover point at 1 kHz instead. I've had excellent results where clients have rarely disagreed with my mastering program equalization. So a large group of people with many different speakers and listening environments tend to like masters that I have produced using that reference loudspeaker system. And it is gratifying to see it resembles the Harman general shape.
5) In the last part of the paper, Olive shows the responses of the loudspeaker equalized to the "preferred target curve" but he did not display the aggregate of that with the initial "preferred" headphone response. I would suggest that Sean supplement his paper to clarify the "preferred" headphone response. The missing figure from Olive's paper would be combination of Figure 17 with Figure 19. So the conclusions of the paper support the conclusion that most listeners prefer the Harman headphone curve shown in figure 17 with increased bass and decreased treble. I can't necessarily agree with the increased bass but I definitely agree with the lowered treble.

In future episodes of Katz's Corner I will produce what the final headphone response curve probably looks like at least with my own ears, after I have done considerable listening with my reference recordings. They should only help to refine Harman's preliminary results which were based on only 3 recordings, any one of which could be too bright, too dull, too bassy, etc. Although I am familiar with two of the recordings they used and these are good starts. But we need a wider variety of selections to see the deviations, to come to a more refined conclusion.

castleofargh's picture

some confusion have been cleared along your successive posts, you got me concerned while reading your first answer ^_^.
yes the harman first target that served as reference for the tests, was aimed to mimic what we should hear from a flat loudspeaker in a room, but inside the headphone. which brings up the problem brought up by Ulrich, that we simply have different bodies so while we can hope to get close to a general curve, there are only little chances for it to be our very own curve.
so we won't exactly hear the same as the flat speakers when listening to the compensated signature inside the headphone. the very basis of the test comes from what is already a general curve only so small differences in the results where to be expected.
to me the next step would be to try and get from people tested, as close as possible their own HRTF and see if the variations align with how they EQed the headphone simulating flat speakers.
that might be the ultimate proof that as doc Olive suspects, we indeed all prefer the sound of a flat speaker playing in a room. which in the ignorant and gullible world of headphones would be a massive break through.

now how reliable are mics stuck inside people's ears to represent the actual HRTF of a person? IDK. as Ulrich mentions, the treble response at least is quite a gamble when doing measurements and minimal changes can bring a few db changes, so audible differences.
and in the end is it really worth it when just placing the headphone differently might also change the signature by a few DBs in the trebles and sometimes in the low end. just like pads aging or people having a bigger head so that the bass seal will be stronger for them on the same headphone? maybe just too many variable right now for a real ideal curve to be done(but I'm still hopeful for the future, maybe scanning our head with MRI to make up a compensation curve or something cheap and simple like that ^_^).

I went pretty far with my so very humble means and IEMs(I never understood why IEMs and headphones are showed with the same compensation curves, but that's another story for another time I guess). and while I'm satisfied with the general direction I got, the same way I'm very satisfied with the general direction of the harman work for headphones, different tips or just slightly changing the insertion depth and I clearly hear another sound. with some resonance that might shift a little or become massive. and that's not accounting for the fact that both my ears have different shapes. ^_^
so it looks to me that we certainly should get a general ideal curve at least as starting point for everybody. at least one that would be subjectively more significant than the actual diffuse field compensation that everybody uses to show measurements while it never sounded neutral to anybody on headphones.
but from there, different people will still hear slightly different things just like with speakers.

to me harman brought what they brought to speakers some years back(thanks for that!!!!), a meaningful work saying "this should be called neutral!". the same way red should be called red. and it doesn't matter that I don't see exactly the same red as you do, we at least all accept that it's called red and we can communicate clearly about it. it's a reference point that hopefully, brands will start to copy so that the overall headphone world can start being concerned about actual specs. and bad products would at last have to stop hiding behind phony interpretations of good or neutral.

also it might not be the most practical place to talk, but doc Olive in the past has come around and was always very gracious to talk to us and clear a few things up when needed. hopefully he will get informed about this all topic going on and come participate again.
not sure about giving us his definitive curve though, as it understandably is a weapon of choice for their future products. ^_^

castleofargh's picture

forgot the part about the bass. the harman test as far as I understood it, didn't really give free rein on the EQ, just one setting for bass with "more" or less". so maybe how much bass people prefer would have changed if the starting point of the increase had been different?
I appreciate about +2 or+3db boost in an almost similar way as showed on tyll's harman make up curve, but it's done on my hd650(from my own poor measurements, not the most accurate and reliable feedback in the world ^_^). the hd650 rolls off a good deal, and I like it with still a little roll off instead of trying to get it dead flat down to 20/25hz. in the end it's more like a little bump in the 40-50hz region to get some rumble fun I miss so much on headphones, than an actual global bass boost. at least that's what I seem to enjoy the most.

Bob Katz's picture

If the constant is the physiology of my ear and the brain accommodates to what comes into that system as natural then it still seems logical to me that matching the eardrum response between loudspeakers and headphones is a good first order approach.

Yes, The variables include measurement accuracy of the dummy head, position of the headphones on that dummy head and differences between that dummy head and actual physiology. But it remains to be heard whether Ulrich's approach will be better or worse in practice.

logscool's picture

Perhaps Tyll could use your DAC test and see what sort of results he comes up with.

markus's picture

Bob Katz's impressions are almost the opposite of the pyrates...

TMRaven's picture

More like the opposite of everybody else who attended big sound. The only headphone everybody can agree on is the HD800.

detlev24's picture

As already stated by others, this is due to different needs, the model of one's brain about what sounds good (personal preferences, linked to habituation) etc.

I prefer a natural music reproduction, as well, adapted to it and I am sure that, e.g., professional musicians (who know their instruments very well) would come to an even more destructive conclusion. ;)

ashutoshp's picture

is that Bob compares them to speakers and only one other participant made a similar comparison. I'm wondering if that's because of the bass responses. In my 2-channel system, I am always floored by the visceral nature of the bass much more so than with headphones where clarity and intimacy are what floor me.

detlev24's picture

that is because of the bass response, but because of the total package (more natural sound reproduction, simplified). Using headphones, of course you miss another important aspect: the bass impact on your body.

On the other hand, you will easily get all the detail that good headphones reveal from a perfectly matched pair of studio monitors. Requirements are for sure an acoustically treated room and the ability to apply further corrections, based on measurements, via DSP: see Bob's working station.

detlev24's picture

A limitation is that most loudspeakers do not extend down to 20 Hz or lower (due to physical aspects). So the use of at least one subwoofer (depending on room size and shape) is recommended; but its implementation can be tricky!