An Acoustic Basis for the Harman Listener Target Curve

Editor's Note: You just never know what's going to happen in my email in-box sometimes.

A couple of weeks ago, I received a very interesting email addressed to Dr. Sean Olive, Acoustic Research Fellow at HARMAN INTERNATIONAL, and myself from Warren TenBrook, an astute enthusiast. Enclosed was a .pdf paper titled "An Acoustic Basis for the Harman Listener Target Curve," in which he describes an alternative way to view the Harman target response. His purpose, was to elicit some feedback on his thoughts, and, since it contained content from both Harman Intl. and InnerFidelity, to ask for permission to publish his paper on headphone forums.

I found his thoughts both elegant and provoking. It's not that there's anything particularly new in the details of his writings, but rather that his viewpoint offers a refreshingly clear way to see the issue from a different perspective.

Evidently, Dr. Olive found his perspective tantalizing as well, and over the course of the next week or so, he and TenBrook exchanged a few emails clarifying the concepts in the paper. What follows is a mash up of the original paper interspersed with further comments by Olive and TenBrook's subsequent back and forth in placed where topically relevant.

Enjoy!

P.S. Please read to the end as I have, what I consider, a very interesting announcement to make!

An Acoustic Basis for the Harman Listener Target Curve
by Warren TenBrook

Dr. Sean Olive and Todd Welti of Harman research have completed several listener testing studies to characterize preferred headphone sound under controlled conditions. They've posted their latest presentation on headphone target curves and listener preferences:

http://seanolive.blogspot.com/2015/11/factors-that-influence-listeners.html

The starting point for this work is a pair of Revel F208 loudspeakers measuring flat on axis in an anechoic chamber with wide, smooth directivity, placed at +/- 30 degree stereo positions in the Harman reference listening room. Depending on set-up, the room acoustics are capable of smooth 0.4-0.5sec RT60 reverberation across the audio spectrum in accordance with ITU-R BS.1116-1. Here is the unequalized speaker curve:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img1

There are normal frequency response variations in the room, particularly in the bass; however, I have added a 1dB/octave trend line to highlight a gentle downward slope. For comparison, here is a graphic of a JBL Synthesis professional room correction, including subwoofers. The JBL target curve and measurements before and after EQ are plotted. Again, the entire curve family slopes 10 dB across the spectrum, 1dB/octave:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img2

It is also instructive to compare the classic Bruel & Kjaer curve from their 1974 AES paper measuring multiple critical listening rooms and studios. There is a 6dB decrease from 160Hz to 20kHz, about 0.9dB/octave, but the pattern is consistent.

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img3

Both Dr. Olive's initial speaker measurements, the JBL Synthesis measurements, and the B&K curve agree that high-quality speakers measuring flat in an anechoic chamber tend to have a ~1dB/octave "room gain" curve when placed in a good-sounding room with no equalization. Also, JBL Synthesis (a Harman company) room correction does not equalize the speaker system to be flat in the room. They smooth the room frequency response to conform to a 1db/octave natural room gain target.

Importantly, the next step in Dr. Olive's headphone listening test method is to equalize the speakers flat in the room:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img4

A Graz head simulator is then used to record the frequency response at the eardrum reference point (green curve):

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img5

Starting with the green headphone EQ curve, listeners were given headphones with ability to adjust bass and treble. In separate experiments, listeners arrived at the black (2013) and blue (2015) curves above, with somewhat more bass and slightly less treble, perceived as accurate or preferred.

This approach suggests two concerns, one technical, another philosophical:

First, Dr. Olive's research uses a flat in-room EQ as a reference point for subsequent listening tests and filter adjustments, which removes the observed 1dB/octave room gain of an accurate loudspeaker in a good room. I would predict that the flat in-room equalized sound would be too bright and bass-light for most listeners versus the unequalized sound of the Revels. Harman's listening test subjects agreed, and preferred increased bass and decreased treble. The flat in-room EQ does not appear intended to optimize the sound of the speaker in the room, but instead seems to provide a raw, blank canvas for the listeners to adjust the sound signature to their preference during the experiment. It is important not to confuse this 'palate cleansing' flat in-room EQ with accurate loudspeaker room response or accurate headphone EQ.

Sean Olive - I appreciate your thoughtful comments on headphone target curves. I think we fundamentally agree on how to derive a headphone target response but there is some confusion on some of the steps we took in our work.

1. I think our target curve from the 2013 paper was indeed based on acoustical principles: it was first derived based on simulating the response measured at the DRP [earDrum Reference Point; as measured by a head simulator at the ear drum] of an accurate loudspeaker (Revel F208) measured in a reference listening room. We showed that the Revel F208 has a reasonably flat smooth, wide-band on-axis response and smooth off-axis sound and sound power based on the anechoic measurements.

We agree that based on previous experiments (also see Toole's recent paper http://www.aes.org/e-lib/browse.cfm?elib=17839 ) that such a loudspeaker will produce good results as long as acoustical interaction between the loudspeaker and room are dealt with below 200-300 Hz, and the bass is adjusted to produce the 10 dB or so downward slope from 20Hz-20kHz. Depending on the room and the loudspeaker-listener, this may require some extra LF gain to produce the rise below 125Hz or so. So before we "flattened" the in-room response of the loudspeaker we essentially had a headphone target response based on an ideal in–room loudspeaker response.

Warren TenBrook responds - Thanks for such a generous reply.

For point 1, I think some of my confusion comes from my interpretation of documents in your blog posts. Your reply mentions you had measured the ideal in-room response. However, I couldn't infer exactly which published curve represented this response. I was under the impression that either flat in-room EQ was being presented, or the listener preference curves.

On closer scrutiny, I see slide 5 of your presentation "Factors that Influence Listeners' Preferred Bass and Treble Balance in Headphones" includes an illustration of a 'Harman Target Curve" that looks virtually identical to my estimate.

Slide 11 also is close, with a bit less bass and slightly higher 3 kHz treble; however, it appears in the presentation after describing the F208 EQ'd flat at the listening position, so I was was not confident this was the 'ideal' in room curve.

Slide 30 is clearly labeled 'Loudspeaker equalized to flat in-room response', so that is the illustration I used to apply 1 dB/octave for my estimate.

If you could point me to the "headphone target response based on an ideal in–room loudspeaker response" my questions would be largely answered. I'd bet on Slide 5, the first graphic below. I'd also be curious if both channels of the stereo pair are driven during measurement, or perhaps summed later, so each ear in the dummy head would include cross-feed from the other channel.

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img10

Sean Olive responds - I'm glad we had this conversation to clarify some of the confusion in our work.

Below is a link to a PDF summarizing the headphone target development. Hopefully it will clarify your questions,

https://dl.dropboxusercontent.com/u/16343460/History%20of%20Harman%20Headphone%20Target%20Curve.pdf

As I explained before (and in these slides) the Harman Headphone Target Response matches the preferred in-room loudspeaker response with some tolerances ( +/-1 dB ) in the preferred bass and treble gain.

The preferred in-room loudspeaker response is a smooth curve from 20 to 20 kHz with about a 9-10 dB downward tilted slope. We agree that a flat in-room response for a loudspeaker will sound too bright and thin. We used that as a baseline for adjustment. In the end, if the loudspeaker has a flat on-axis response you shouldn't need to adjust the treble, but the bass will probably need adjustment depending on the room, and the positioning of the loudspeaker and listener.

Todd Welti is working on an explanation of how he measured the loudspeakers in the room. But essentially he used a GRAS 45 flush mounted to a head and measured each loudspeaker channel independently at the DRP with head in the sweet spot. He assumed symmetry for the head/ear and in the room and loudspeaker setup so the left ear signal for the left loudspeaker would the same as the right loudspeaker measured at the right ear. I think this is fair based on the shape of the room and the loudspeaker setup, which was symmetrical. He did a spatial average for each measured channel for +/-30 degrees and 0 degrees.

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img11

Sean Olive - 2. What seems to bother you is that we flattened the in-room response as a sort of baseline curve for the method of adjustments. It was not our intention that this flattened curve be a "reference" that listeners could hear and compare their adjustments to. The flattened curve was only a baseline from which the bass and treble adjustments were made. We never provided listeners this flattened condition as a "reference" in the listening experiments. In fact, the starting bass and treble levels were randomized in each trial so that there could be no bias effect related the starting levels. An important note is that original in-room loudspeaker response (before flattening) was one of the options that listeners could choose by adjusting the bass and treble controls to the appropriate settings. In the end, listeners preferred a headphone target response that came pretty close to the preferred in-room target response of the loudspeaker using the same programs.

In the 2013, paper they preferred slightly less bass than the loudspeaker. In the more recent paper, the average preferred bass level was the same.

Second, I support listener preference research, but I feel that subjective listener data using two broad bass and treble filter adjustments should remain focused on documenting listener preferences alone. I don't support extrapolating these subjective preferences to broadly-applicable target curves for design, testing, or purchasing headphones as this contributes to what Dr. Floyd Toole dubbed 'The Circle of Confusion'. In my opinion, a measurement specification like a headphone target curve should be derived from objective physical and acoustic principles; listening tests can then inform us if the objective curve is reasonable and consistent with preferences, i.e. "musical".

Sean Olive - 3. Your second argument, is that preference experiments should NOT be used to define target responses for headphones, in part, because of circle-of-confusions issues. All of the loudspeaker research (Toole @ Olive) and headphone research (Olive & Welti) have involved preference ratings to help support acoustical hypotheses of what the ideal headphone and loudspeaker should look like in terms of acoustical measurements.

Whenever you do listening tests you have to use music programs and cannot avoid the circle-of-confusion issues. So you try to choose neutral, sensitive programs and use many programs and use statistical averaging. Ideally, we should have used more programs statistically sampled from what people purchase but we don't have the time and resources to do so. We acknowledge this in the paper and say that the ideal target response is a moving one that will depend on the recordings you listen to.

If the professional recording industry used calibrated, standardized monitors that were well-defined acoustically to record and master music, then it would be simple to define the target response of headphones and loudspeakers: you simply copy their standard. But because no such standard exists we have to define a statistical average of what they use and hope the recordings were made using that average.

The good news is that statistical average of most professional monitors is probably close to something like a Revel F208 (flat on-axis, smooth sound power) or a JBL LSR632 or M2, which means our headphone target response is pretty close. What we did find from this research is that the preferred bass and treble levels depends on the age, listening experience and gender of the listener. Younger people with less experience preferred more bass and treble. Older people preferred more treble (to compensate for hearing loss?). This is helpful in explaining why some listeners may not prefer a headphone or loudspeaker that is preferred by others.

Warren TenBrook responds - Point 3 is most interesting. Again, when I could not readily pick out the measured 'ideal in-room loudspeaker response' curve, I began to almost view the listener test data as obscuring what I wanted to learn about the in-room measurement, and that led me to imply that you were largely basing the target on listener data, and de-emphasizing the in-room measurements which were my primary interest. I wanted to convey my opinion that listening tests are not an end of themselves, but a confirmatory tool to make sure the technical approach holds water.

Then I read your insightful reply "...If the professional recording industry used calibrated, standardized monitors that were well-defined acoustically to record and master music then it would be simple to define the target response of headphones and loudspeakers: you simply copy their standard. But because no such standard exists we have to define a statistical average of what they use and hope the recordings were made using that average..." This caused me to recall the work at NRC on listener testing of loudspeaker characteristics. Dr. Toole and other's work to find correlations between measurements with listener preference changed the industry. (I still have my trusty Paradigm 5se bookshelves from years ago. I think their design was based on Floyd Toole's work—and they still sound great). My impression is we are getting close to a de facto speaker standard building on this research. But it never could have happened without listener tests to guide what measurements were relevant to good design. Clearly this is important to a headphone target as well.

Even without a generally-accepted standard for mastering studios—and assuming a majority of recordings weren't mastered on accurate speakers in a good room—I'm glad Harman is putting a stake in the ground and proposing that its speaker products and reference room reflect state-of-the art and are suitable to derive the headphone target curve.

Sean Olive - 4. In conclusion, I believe our listening tests confirm that the preferred headphone target response closely matches the ideal in-room target response of a loudspeaker measured at the DRP.

I would argue we defined the curve based on acoustic principles (measuring an anechoically flat speaker calibrated in a reference listening room) and confirming that people like it through a method of adjustment listening test. In the listening experiments, the green curve (flattened in room curve) was only one of many options including the ideal 10 dB-octave sloped in-room curve. It was not a "reference" that listeners could access as an anchor but rather one of many options within the limits of adjustment. One of the adjustment options was the so-called ideal in-room target curve.

In the end, the listening test data shows that listeners rejected the flat in-room curve, which has been shown in previous studies, and preferred something that resembles something you propose. You could go through the exercise of measuring many other speakers in many different rooms to acoustically derive a new target curve. But if you look at all the studies together (see Toole's paper), there seems to be consensus on what the ideal in-room loudspeaker response should be: in rooms, cars, cinemas and hopefully soon, headphones too.

An Acoustic Basis for the Harman Listener Target Curve (continued)
I wondered if there was a physical/acoustic approach that would favorably compare with the Harman listener test preferences. What would happen if I restored the 1dB/octave room gain slope to Dr. Olive's measured flat in-room 'green curve'? Here is the result:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img6

The adjusted curve is similar to the Harman listener curves in key respects:

  • The adjusted curve has a gentle rising midrange from 200 hz to 3000 hz.
  • Treble is tamed compared to the flat in-room speaker measurement and returns to baseline level at 10 kHz.
  • The 1 dB/octave room gain bass rise from 200 Hz to 20 Hz is similar to the Harman listener curve, without a bass 'shelf' under 100 Hz. The 'shelf' might be related to the bass filter shape used for the Harman studies (another reason not to derive target curves from the two filters used in the listening tests).
  • Tyll Hertsens has commented on bass emphasis in several headphones bleeding toward the 200+ Hz range can lead to upper bass and lower midrange thickness that is not neutral. The adjusted curve backs off the upper bass level slightly more than the Harman curves, particularly by eliminating the bass shelf.

Here is a comparison of the curve with some sample headphones. First, the Oppo PM-3 follows the curve very closely, with treble somewhat rolled-off as observed by Bob Katz:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img7

Next, the Ether C is slightly under the curve in the bass and has a bit more energy at 3-4 kHz, but the midrange and treble follow the curve very closely:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img8

Sennheiser HD650, an open headphone, follows the curve well, with a bit less bass extension and slight midbass warmth:

/images/160503_Blog_AcousticBasisHarmanTargetCurve_Photo_Img9

It appears that measuring accurate loudspeakers in a room with good acoustics and natural 1dB/octave room gain would provide a measured target curve close to that preferred by Harman listener tests. I predict an accurate objective target curve for headphones could be derived from:

  • Rooms with controlled reverberation time meeting ITU specifications. Testing in different rooms meeting the specification would be beneficial.
  • Speaker placement at +/-30 degree standard stereo.
  • Relative speaker distance to room boundaries and to listener position selected to approach smooth room response from the stereo pair.
  • EQ, if used, should be limited to smoothing the natural room gain of the speaker, but not force the speaker to be flat at the listening position. A flat EQ at the listening position is too bright. Instead, room gain of approximately 1dB/octave downward slope is expected.
  • Testing with different artificial head/torso apparatus meeting IEC specifications would be beneficial to bracket the range of potential measurement variations.
  • A profile of several well-designed conventional dynamic loudspeakers is preferred, representing mastering studio quality (anechoic flat with wide, controlled directivity). Each good speaker will have somewhat different directivity that will offer a range of potential interactions in a good room. Examples might include:

    • JBL M2 Studio Monitor
    • KEF Reference Series
    • Revel Salon 2
    • Dynaudio Evidence
    • TAD Reference One

By performing various measurements with the same general approach to room acoustics, selection of reference-quality speakers and simulated head measurement techniques, a family of curves could be presented, including a best fit average curve. One convenient approach could be to engage several well-engineered mastering studios and do multiple torso?head measurements in each.

I predict the resulting curves would be similar to the Harman listener preferences and would be an excellent technical definition of a neutral headphone curve.

Editor's Note: First, thank you Warren for permitting me to publish this work-in-progress. I look forward to your further thoughts on this subject, but ultimately felt there was so much good information here that publishing your thoughts, and Sean's responses as is was a worthy endeavor.

Further, Warren has pushed me over the edge. I've been, for a long, long time, very curious about what a good set of speakers in a good room would look like when measured by my head. As part of my effort to take a good hard look at artificial head measurements, I asked Sean Olive if it would be possible for me to visit Harman's research facility with my gear and take some measurements with my head. He has graciously replied that I could do so. I'm stoked! The measurement session will be in early June. Woot!

COMMENTS
GNagus's picture

If you don't mind paying for shipping I can send you my Bose 301, Series III I think (the ones with the double tweeters, one pointing one way and the other pointing the other way) to measure so that you don't have to go nowhere.

detlev24's picture

WOW, that is very generous of Dr. Olive! Props to Harman. I hope you get to listen to the exceptionally good M2 studio monitors. :D

Let us hope you will be allowed to post some photos of your visit.

THX for this article!

Best regards

wktenbrook's picture

Yes, this was just an innocent submittal to make sure use of the illustrations wouldn't cross anyone, but it turned out to be a fascinating conversation. I feel very fortunate.

dsnyder0cnn's picture

Thanks for sharing this. Finding a great sounding target curve has been a preoccupation of mine since I started experimenting with room correction this past December. I typed up some my thoughts, along with some practical resources in a blog post called Can Digital Make a Room Correct?. Not as authoritative as the discussion here, but I hope it's helpful to some.

ADU's picture

Interesting read. Thanks for posting this, dsnyder0cnn.

Johan B's picture

Tyll,
Great article but with all probabilities of going down the slope :-) In the living room speaker set up does Harman really turn the speaker center towards the ear? If they keep the speaker straight the the listener will hear the "off center" response that has indeed a frequency response decay towards high. This is however a problem that manufacturers are trying to correct and so Stereophile measure the response as optimal or good if there is no decay. So to say that the "normal" response assumption in room from Harman may not be "high end". A good example is that studio monitor speakers are listened to up close and center to ear. These do not suffer from the "in room" decay response and are often describes as "neutral" and accurate. Note that most music is mixed that way. So all in all I am a bit in doubt about where this is going?

Johan B's picture

http://www.stereophile.com/content/kef-blade-two-loudspeaker-measurement...

The speaker with perfect in room response is ... linear.

detlev24's picture

In room response depends from the room, for every(!) sound source. You will never find a workspace with linear frequency response - there will always be EQ to the B&K 1974 curve or similar (HARMAN, Bob Katz etc.) in the end.

Acoustics is very complex, but be sure that the first thing to do is to treat/build a room very well and only thereafter to apply EQ as needed. This way you can get quite good frequency response off-axis, as well. Especially with a set of speakers like the M2 (solely the loud PA system amplifier(s) are not the best choice for living environments^^).

Of course, there is a sweet spot but it can be smaller or wider - depending also on further details such as loudspeaker placement and orientation. You may find some information on Harman's approach here: http://www.jblpro.com/ProductAttachments/M2_Brochure_Jan2013.pdf

Best regards

wktenbrook's picture

I'd give Olive and Welti the benefit of the doubt here. Although recordings are often initially monitored on near field speakers while studio work is in-progress, it's my impression that the mastering studio is where the final sound signature is decided, and it typically resembles a high end home listening room. Bob Katz could clarify this.

Tonmeister's picture

We also did a spatial average of each measured loudspeaker at the listening seat based on three head positions so that the target curve is not so biased towards a single measurement or HRTF.

With typical forward facing loudspeakers, the directivity increases with frequency. What that means is that at higher frequencies the direct sound dominates what the listener hears, making the room acoustics less of a factor. So whether you are in the Harman Reference Room or a near-field recording control room situation, you are hearing much the same sound above a certain frequency.

At lower frequencies, the room acoustics do make a difference as illustrated in the Toole AES paper linked in the article (it's open access or free BTW). The in-room steady-state response of the loudspeaker below 500 Hz can vary significantly depending on the room dimensions, absorption characteristics, loudspeaker-listener position(s),etc. I'm not sure sitting closer to the speaker is going to fix the "bass problems." A popular method to fix the bass problem in recording studios was to eliminate the bass in the loudspeaker which partially explains the popularity of Auratone 5C and Yamaha NS10's as "reference loudspeakers" ... this is not a good solution since the engineer cannot hear the proper spectral balance going to tape or disc.

Fortunately, there are better solutions today for dealing with the interaction between loudspeakers and rooms acoustics, and these were incorporated when deriving the Harman Headphone Target Response.

Three Toes of Fury's picture

I love articles like this. Thank you so much Warren for taking the time to write and share with us all. Thank you Tyll for posting!

Waveform analysis is continually a tricky concept for me to grasp and i very much enjoy each new take and approach to the topic.

Tyll...i cannot wait to hear the results of your Harman visit!! Grab as much data, photos, opinions, ideas, concepts, comparisons, and any thing else you can and report back to we readers. Should be a hoot!

Peace .n. Living in Stereo

Three Toes of Fury

ericw's picture

From what I can see, agreeing with Warren, the actual target is the dashed line on the graph marked "Harman Target Curve (2013) / HP1" on page 13 of History of Harman Headphone Target Curve.pdf.

This is not the same as either the solid black (listener testing used to validate the target) or dashed green curve ("baseline" with speakers equalized to flat in-room response) on page 10, also seen here: www.innerfidelity.com/content/first-test-estimated-harman-target-respons...

wktenbrook's picture

Agreed. My main point in the write-up was to try to nail down the in-room measurement by other means, since I was admittedly confused by so many curves presented in these papers, all bearing a close family resemblance (which is a good thing because it means listeners preferred the same curve generated by the in-room measurements, under double blind conditions with complete freedom to adjust bass and treble levels).

ADU's picture

I'm still (mostly) at Audeze treble response, :) but can see several potential pitfalls with Warren's room correction approach above.

1) There isn't really a typical in-room response for loudspeakers. Based on the data in some of the links above, the in-room or "steady-state" response is dependent on the speaker's position (and alignment as Johan mentioned), the size of the room, and also the reflectivity of the room's surfaces (walls, ceiling, floors, etc.). This is really the crux of Floyd Toole's work, and his contention that EQ-ing to one specific in-room response curve is the wrong approach. The main goal for _music_ loudspeakers is still a flat direct/anechoic response, and reasonably smooth (though not necessarily linear) in-room response.

There are similarities between the measured and predicted responses for different rooms though, which brings me to potential pitfall number two...

2) This may be splitting hairs to some, but recent predictions, measurements and subjective preferences for in-room loudspeaker response do not generally seem to follow a constant slope of 1 dB per octave. Though there does seem to be an overall drop on the general order of 10 dBs (give or take a few dBs) from the lower bass to the upper treble (20 Hz - 20 kHz).

Generally speaking, the shape of current in-room curves tends to be more serpentine, with a bump in the bass, a somewhat flatter or slightly inclined area in the mid-range and/or low treble, and some degree of roll-off at the higher frequencies in the treble. (The curves are somewhat open to interpretation though. So others may disagree with this characterization.)

http://www.aes.org/tmpFiles/elib/20160504/17839.pdf

Floyd covers alot of ground in this article (and spends alot of time explaining why the "X-curve" is wrong). So I'll point to some of the graphs that I think are most relevant here for music listening. The heavy dashed lines on Figures 22a and 22b show the predicted in-room response for larger venues with average room reflectivity. Both have a boost in the bass, a flatter area in the midrange and roll-off in the upper treble.

Figure 13a shows the predicted in-room responses of 3 good speakers based on anechoic ("spinorama") directivity data, which is represented by the squiggly lines on the first curve. And the squiggly lines on Figure 13b show the actual measured response of those speakers in a small venue, such as a home theater. The rise from the midrange to bass is more gradual in smaller venues. And there's not quite as much roll-off in the upper treble, possibly because there's less distance and air to absorb the higher frequencies between the source and listener. But these also follow the general trend of the other predicted room curves with a boost in the bass, roll-off in the treble, and a flatter area somewhere in-between (which is more in the low treble in this case).

Figure 14 shows the preferred in-room (music) response for untrained listeners (dotted line), and trained listeners (thin dashed line). The untrained listeners prefer more of a "smile" with a strong boost in the bass and flat or slightly upward tilt in the treble, while the trained listeners prefer a more gradual descent from bass to treble. Both preferred _some_ bass boost though consistent with the other in-room responses mentioned above, and also a degree of roll-off in the upper treble.

Figures 19a and 19b show the in-room measurements of the reference JBL M2 speaker for various larger venues, with microphones at different positions. Once again, the general trend is a boost in the bass, flatter midrange, and rolled-off treble. (The occasional dips well below 0 dBs in the bass and lower midrange are likely due to interference from the theater seats.)

Car audio systems follow this same general pattern...

http://i12.photobucket.com/albums/a224/pionkej/Murano%20Build%20Log/Targ...

The average target for _in-car_ audio, which is outlined by the orange areas on the above graph, has a bump in the bass, a flatter slightly inclined area in the midrange, and a rolled-off area in the high treble, just like the latest in-room predictions and measurements for loudspeakers. And the overall change in amplitude from 20 Hz to 20 kHz is also on the order of 10 dBs.

This data is also relevant imo because listening to music in the smaller more confined space of a car with speakers closer to your ears is closer to the experience of a pair of headphones than loudspeakers in a room (others may disagree though).

If you prefer to simplify the room response to a 1 dB per octave slope, that's certainly your prerogative. But it doesn't accurately reflect the current data imo, based on all of the above info.

3) Warren also neglects to take into account Tyll's EQ corrections on his raw headphone FR plots.

4) Despite all the above info, not everyone agrees that room correction is needed or appropriate for headphones. I tend to feel that some correction is necessary to better approximate what the artists and engineers heard, and intended others to hear during mastering. But this also begs another question...

5) If you agree that some room correction is necessary in a headphone, then what room do you try to approximate? Do you try to match the space where the content was mastered, or the space where you're listening to the headphones, or possibly some other fictitious space of your own choice?

If your goal is to preserve the artistic intent of the music producers, then it should probably be the mastering space. A correction based on the mastering space could possibly sound "out of place" though in a different listening environment. And I'm not sure how you'd compensate for that.

5b) The standards for mastering music and movies are also different. If you're watching a movie that was mastered in a room that conforms to the _X-curve_, which looks something like this...

http://www.avsforum.com/forum/attachment.php?attachmentid=263194

...then do you use that as your headphone room curve, or use one of the "music" room curves above, or a curve that matches your listening space, or some kind of compromise between some or all of these options? If you answered "D", then perhaps that compromise curve is closer to a 1 dB/octave slope due to the differences in bass response between the X curve and "music" room curves.

Those are just a few of the factors that I think you'd need to take into consideration when deciding what room correction to use for a headphone. And there are undoubtedly others as well. If you're using an EQ to apply the necessary adjustments to your headphone's overall frequency response, then you might want to target response curves based on more than one type of room response,... maybe one for music and another for movies. Given the above info that would seem to make alot of sense.

The general message that seems to be coming from the acoustic data (and Tyll's HP reviews) though is that some degree of boost (or "shelving" if you like) in the bass and roll-off in the treble is probably appropriate for music listening. And the type of boost and roll-off could be either gradual or more pronounced, depending on the size/type of room or mastering environment you're attempting to approximate.

I've only skimmed Floyd Toole's PDF referenced above btw, but have been looking at room response data and targets from various sources since rewatching his presentation for the CIRMMT in an attempt to figure out how to better adapt Diffuse Field curves to a neutral response.

https://www.youtube.com/watch?v=zrpUDuUtxPM

I think the Diffuse Field functions are going to need a stronger correction though, esp. in the treble, to compensate for certain directivity issues.

wktenbrook's picture

Great post, ADU.

I agree there is no 'typical' response. The best I could suggest was measuring several different good rooms (ITU spec), with several different respected speakers (good anechoic measurements/directivity) and calibrated heads (IEC spec), then look at the data spread and trend lines to bound the variation. A neutral stereophonic playback headphone curve would be somewhere in those weeds.

In-room calibration is full of pitfalls. I think Harman Research philosophy is to EQ below the standing wave transition frequency and leave well enough alone above 200-300 hz if the speaker is well-designed. This approach can minimize the temptation to slavishly apply room EQ curves that could make matters worse.

I cringe when I hear about different approaches for music versus movies - headphones have but one channel for each ear after all - but then I saw that Harman's headphone research has measured both multichannel and stereo playback. I have to think the HRTFs are altered with the addition of more channels and the curve is different, even if I'd prefer it to be simple and consistent.

I suppose with enough sound sources, like immersive audio, you might start to approach diffuse field!

Tonmeister's picture

Hi Warren:

Our conversation continues.. Actually, the differences measured at the ear drum between the Stereo versus Multichannel Loudspeaker setups were negligible. This is probably because we used well-matched speakers in an ITU-setup (all equal-distant from the listener, at the same elevation, in a symmetrical room) and four bass-managed subwoofers to produce a very smooth response across the seats. The in-room measurements of the 7 loudspeakers in the paper illustrate how similar they were. So the bass and direct sound was not much different from each channel.
So the measured differences at the DRP would be mostly related to the angle of incidence of sound arrival, which affects the HRTF above 1-2 kHz. When the 7 loudspeakers are measured using a spatial average of 3 head positions, the response at the ear drum was not significantly different between stereo and multichannel conditions.

detlev24's picture

Please allow the following question. What is the point of measuring at the DRP - with regards to over-the-ear headphones; instead of measuring at a neutral reference point (not yet influenced by human anatomy) and try to simulate the acoustic space we expect from loudspeakers, before it hits the human body and thus frequency response gets altered individually at the ear drum?

I understand this approach with regards to IEMs but I cannot follow in case of over-the-ear headphones. Thank you!

Best regards

ADU's picture

Many thanks for the compliment, Warren.

Quote:

I suppose with enough sound sources, like immersive audio, you might start to approach diffuse field!

The reason I'm still (somewhat) interested in the Diffuse Field is because it does a better job of modeling the HATS resonances at approximately 3, 9, and 15 kHz in the treble than the Harman Curve. The Diffuse Field is the dashed line on this graph...

http://en.goldenears.net/en/files/attach/images/249/388/e933afabab8f22b9...

Before going into that though, I want to make a few corrections/additions to some of the info in my last post. First thing I should do is correct the bad link to Floyd Toole's PDF. One of these should work...

http://www.aes.org/e-lib/download.cfm/17839.pdf?ID=17839

http://www.aes.org/e-lib/browse.cfm?elib=17839

I'm still working my way through this article. But I think Floyd does a pretty good job summarizing the dilemma with in-room (aka steady-state) response curves at the end of page 515 and beginning of 516...

Quote:

There is a difference between the spectrum of the direct sound arriving at a listener and that of the steady-state sound level that is achieved after reflected sounds arrive. The shape of a steady-state room curve is determined by the sound radiated by the loudspeaker modified by the geometry and frequency-dependent reflectivity of the room. In an acoustically dead room, the room curve will be identical to the on-axis response of the loudspeaker. As reflections within the room increase, the room curve will rise towards the predicted room curve, as the offaxis sounds add to the result. The bass and midrange sound levels will build up over a short time interval, affecting what is measured and heard. At very high frequencies the direct sound becomes progressively dominant. Therefore with no knowledge of the loudspeaker, and no knowledge of the room acoustical properties, a steady-state room curve conveys ambiguous information.

So room curve = ambiguous info. And that's what we're trying to model on our headphones. :)

Anyway, I think what he's saying is that the sounds you hear in a room from a loudspeaker are not just a homogeneous wall of frequency information that can easily be transplanted from one place to the next via a generic room curve. And even if you could do something like that, it would probably still sound wrong, because every room has its own unique frequency response, which is dictated by its size, reflectivity, etc.

So the best that most of us can do is get a speaker with a flat direct/anechoic response (and reasonably smooth, distortion-free off-axis response), put it in a room, and let it do its business. If the music/sound engineers have done their job correctly on the recording and production side, then you should experience something pretty close to what they did when the music was mastered and created.

The only way you could conceivably enhance that experience is by building a room that's the same size and shape as a typical mastering suite (if such a thing exists), and give it a similar acoustical treatment. It's really the sound engineer's job though to do his best to match the typical _untreated_ acoustics in your home, rather than the other way around.

Movies could be an exception to this rule, because they're often mastered to a pre-defined room spec (namely, the X-curve), which may or may not approximate the untreated acoustics in your home. (It sounds like this could be changing though, due to the efforts of Floyd, Sean and others.)

Maybe someday there'll be a way to analyze _all_ the sound info in a room, and be able to replicate that inside a pair of headphones. I wouldn't put that past some of the brighter technologists out there. Until then though, we are basically stuck trying to do exactly what Floyd and others advise against with loudspeakers. Namely, transplanting the sound of a room into a pair of headphones via a simple generic room curve.

wktenbrook's picture

“So room curve = ambiguous info. And that's what we're trying to model on our headphones.”

ADU, I think Dr. Toole readily admits that divergent room acoustics can impart room to room differences to the listener even with an well-designed speaker. But his work has concluded that a speaker that is anechoic flat on axis with controlled directivity will best adapt to various rooms. This allows for some “room EQ curve” adjustment at the listening position while keeping both the on axis and reflected sound somewhat consistent. I think your quote means that if the room or speaker are not up to spec, there is no way to correct the flaws by doing EQ at the listening position.

“Therefore with no knowledge of the loudspeaker, and no knowledge of the room acoustical properties, a steady-state room curve conveys ambiguous information.”

Toole does not say all room curves are ambiguous, but room curves without information about the underlying acoustics, speaker playback, listener position will be ambiguous. So we have to approach the room, speaker and listener as a system and decide on a neutral baseline spec for each aspect. We have a de facto spec for neutral speakers that several respected brands are approaching. ITU specs define a reference for listening room acoustic, and Harman's reference room is capable of meeting the spec. Dr Olive stated their stereo and multichannel measurement were reasonably consistent. There a bit of intrigue between Tyll and Audeze head measurements, but I’m not willing to concede that head measurement systems are all over the map. So I’m encouraged we can limit the variables underlying a neutral headphone curve.

There is still plenty of room to argue if the industry specs adequately characterize a neutral playback chain, and how relevant the specs are to the way typical listeners play back recordings, but we have to start somewhere. Again, Harman's listening panels converged on the good speaker-good room ideal curve even with randomized bass and treble starting points under double blind conditions. This suggests that we can transplant at least a facsimile of the sound of a room into a pair of headphones via a simple generic room curve.

ADU's picture

Another thing I wanted to come back to briefly was the comparison of different audio targets for cars.

The graph in my last post came from here...

http://www.diymobileaudio.com/forum/technical-advanced-car-audio-discuss...

There's a similar graph on page 529 of Floyd's Toole's PDF in Figure 15...

http://www.aes.org/e-lib/download.cfm/17839.pdf?ID=17839
http://www.aes.org/e-lib/browse.cfm?elib=17839

I believe the Olive & Welti target shown on that graph is the same as the blue JBL curve on this graph...

http://i12.photobucket.com/albums/a224/pionkej/Murano%20Build%20Log/Targ...

The only difference is some smoothing and a bit more tilt in the mid-range in the Olive-Welti plot.

The Olive-Welti/JBL target has the same general "serpentine" shape as the average of the 5 car systems, with a boost in the bass, flatter slightly inclined (toward the bass) midrange, and rolled off treble (which is also the same general shape as the in-room curves for music indicated in my last post). The main difference is the bass level, which is 4-5 dBs higher on the Olive-Welti/JBL target than the average indicated by the orange shapes.

The extra gain in the bass on the Olive-Welti/JBL curve is apparently to compensate for low frequency road and engine noise masking some of the bass information from the car's speakers. Presumably, a better insulated car would require less of a boost to compensate for that external noise.

YMMV, but I think the average car curve (without the extra JBL bass boost) outlined in orange would probably make a better room target for a pair of neutral headphones, because it's a little closer to the ~10 dB overall slope convention for other room curves of this type.

Maybe a little more tilt in the midrange would be warranted to compensate for the lack of smoothing on a few of the curves. I'm not really sure about that though.

wktenbrook's picture

I haven't explored the science behind car audio. All I've ever done is audition my car playback and immediately compare with headphones I trust, then adjust the car EQ to match as best I could.

At first I thought car audio should be easier for manufacturers to get right since the listening positions and environment are well-known, but now I have doubts. There are so many heterogenous materials, angles, reflections, masking noise that is seems a very difficult environment compared with headphones or listening rooms. I would like to see what would happen if you adjusted an automotive system to give the same response on a HATS In the drivers seat as in a good speaker-room setup. I have a hunch Harman is already doing this to support their JBL and Levinson car systems.

ADU's picture

There are several issues I want to touch on re the current Harman headphone curve...

https://dl.dropboxusercontent.com/u/16343460/AES%20139%20Presentation%20...

The first is this comment on page/slide 33 of the above PDF...

Quote:

Method of Adjustments in bass and treble not loudness compensated: some listeners may have simply boosted bass and treble because it made the music louder

This seems possible to me. Because I think people generally tend to prefer more bass and treble at lower listening levels. And the shape of the average Bass & Treble Filters Gain curve on page/slide 19 appears to fit somewhere between the average preferred in-room response and the more "smiley" untrained preference shown on page 528 of Floyd Toole's PDF in Figure 14...

http://www.aes.org/e-lib/download.cfm/17839.pdf?ID=17839
http://www.aes.org/e-lib/browse.cfm?elib=17839

The Harman Target test subjects could also be overcompensating the bass and treble adjustments simply because there are _too_few_ controls to really shape the headphone's sound the way they want.

The Bass & Treble Filters Gain curve on page/slide 19 of the Harman Target PDF (1st link above) doesn't precisely match the serpentine shape of the other room (and car cabin) curves described in my previous posts. I suspect that's because _both_ the bass and treble controls may have too much of a shelving effect, and there's not enough control over the slope of the rise in the bass, or the roll-off in the treble, for example.

Maybe an overall volume control, and a few more points of control along the frequency spectrum (or a different type of bass and treble adjustment that produces less shelving) might yield results that fit better with the measured and predicted in-room curves in Floyd's PDF, when averaged across the test subjects. Until then, I think the current Harman Bass & Treble Filters Gain curve and the overall levels in the Harman Headphone Target should be taken with a little grain of salt due to inconsistencies with other in-room steady-state predictions and results.

My biggest issue with the Harman Headphone Target though is still its poor modeling of the HATS (head & torso simulator) resonances in the treble, which regularly appear in average plots of the better headphones in both the Inner Fidelity and Golden Ears databases.

I've painstakingly plotted (by hand, a point at time) the raw frequency response of around 40 headphones in both the Golden Ears and Inner Fidelity databases on my computer, and calculated the averages for various groups of those headphones. And the results are virtually always the same in the treble. There's always a series of peaks or "bumps" at around 3, 9 and 15 kHz which look something like this...

http://www.avsforum.com/forum/attachment.php?attachmentid=922226

(Note: the main driver resonance in the bass is also indicated on this graph.)

The more (good) headphones that are averaged together, the more obvious and well-defined the three peaks in the treble become.

Roughly the same pattern of resonances at 3, 9 and 15 kHz can be seen in the Diffuse Field plot of the HATS system used by Golden Ears, which is represented by the dashed line on this graph...

http://en.goldenears.net/en/files/attach/images/249/388/e933afabab8f22b9...

That could be just a coincidence. But the more logical explanation is that the resonances at 3, 9 and 15 kHz are a feature of the simulator rather than the headphones, and not something that should simply be "corrected out" by targeting a smoother response curve like the Harman Target...

http://cdn.innerfidelity.com/images/160503_Blog_AcousticBasisHarmanTarge...

The problem with the Diffuse Field curve is that it doesn't model any of the aforementioned room or directivity effects that you'd normally get when listening to loudspeakers set at a 30 degree angle in your home or a mastering environment. And that makes a _huge_ difference in the sound, esp. in the treble, because there's no absorption, beaming or angle of incidence to the listener taken into account.

From purely a levels standpoint, the Diffuse Field curve dramatically over-shoots the in-room brightness of a loudspeaker at the highest frequencies, and under-shoots the bass response, making it a poor model for a neutral headphone response.

Golden Ears' solution to that problem is to apply both an X-curve correction (discussed in my previous posts) and a bass boost to the Diffuse Field curve, and use that as their target headphone compensation curve. The result is the dashed line on this raw FR plot of the Audio Technica R70x...

http://ko.goldenears.net/board/files/attach/images/783981/808/845/005/c1...

Their approach seems to be based mostly on guesswork though, rather than acoustic measurements, which makes their compensated results "suspect" imo, esp. in the upper treble response...

http://ko.goldenears.net/board/files/attach/images/783981/808/845/005/f9...

At least the GE target curve includes the resonances at 3, 9 and 15 kHz though (from the Diffuse Field curve). Imo, these resonances should be incorporated into the target curves (and hence compensated plots) of _both_ the GE and Inner Fidelity headphone measurements, because they are features inherent to both measuring systems. And omitting them from the target curve would, imo, lead to a less reliable/accurate compensated result.

The one sidenote I'd add is that the resonances don't always occur at precisely the same frequencies in different measuring systems. So the target response curves need to be "tailored" a little for each system. In the Golden Ears plots, for example, the main treble resonance tends to peak at around 2.7 kHz. While in the Inner Fidelity plots it's a bit closer to 3.5 kHz.

After giving this problem quite a bit of thought, I think I may have figured out one potential _non-acoustical_ way to make these kinds of adjustments without significantly compromising the accuracy or integrity of the target curves. And I may post a few examples of this when I get a chance.

I think the "room correction" component in the new Harman Headphone Target probably still needs some polishing though, to bring it into a little better sync with the latest in-room results for loudspeakers shown in Floyd's PDF, esp. in the upper treble.

The average frequency response plot posted above is based on 21 of the better headphones from different mfrs. in the GE database btw...

http://goldenears.net/board/GR_Headphones

I plotted 81 points (at 1/6 octave intervals in the bass/midrange and 1/12 octave intervals in the treble), for both the left and right channels of each headphone, which are represented by the white dots on this graph...

http://www.avsforum.com/forum/attachment.php?attachmentid=922218

And then averaged the points together at each interval to create the average curve in green. I think that's a total of 3402 plotted points. :)

I used twice as many points in the treble to more accurately model the fluctuations in that area. And the curves were all centered at an arbitrary point in the middle of the frequency range at 630 Hz, so that's why the point distribution gets progressively narrower towards that one particular spot.

About half the headphones in the sampling were on the brighter side, so the net result may be a little bright of neutral. The AT R70X (which was not included in this sampling) is one the better matches in the GE database to this average curve and also to some more neutral plots I've done since the above curve was produced.

It's a bit easier to make plots based on the GE graphs btw, because the curves are smoother, and there's no compensation to undo on the left and right channels. To accurately plot the raw Inner Fidelity curves, I first have to undo the Independent of Direction correction on the left/right compensated plots, and use 1/24 octave data points in the upper treble to adequately model the unsmoothed fluctuations there. So there are a couple more steps involved. (Undoing the IoD compensation is the most laborious.)

If you fit curves to both the peaks and valleys of the above average plot, the result is a graceful arcing shape that curves smoothly downward in the treble...

http://www.avsforum.com/forum/attachment.php?attachmentid=922242

For those who may still be having trouble following some of the info and graphs above, I recommend giving this presentation and PDF by Mr. Hertsens a look...

https://www.youtube.com/watch?v=SDRHFNfFCFU

http://www.innerfidelity.com/images/InterpretingHeadphoneMeasurements.pdf

Also, images of the averaged headphone plots, including the list of 21 sampled headphones, can be found here if the links above don't work...

http://www.avsforum.com/forum/378-headphones/1530412-what-current-gold-s...

I look forward to hearing more about Tyll's adventures in the Harman reference room btw. In-ear measurements from a room that conforms to the X-curve would also be interesting for comparison.

wktenbrook's picture

I can't draw any clear inferences about test subjects dialing the method of adjustment merely to change the overall sound level to their preference. If they adjusted bass and treble motivated by a 'louder is better' instinct, I should think they would have increased both bass and treble and obtained a smile curve. Instead they tended to reduce treble and results conformed to in room measurements.

The modeling of HATS resonances is a matter deserving attention. My inclination is to call everything above 3 kHz up for grabs since every person's auditory system will exhibit different resonances, and different measurement heads may differ as well, so best to simply use a smoother curve that captures the overall trend of typical hearing, not sweat the details above 3 kHz, and let each listener's individual auditory resonances fill in the details during headphone listening.

You've compiled a great deal of data suggesting there are key resonances above 3k. If these resonances are common to particular types of HATS equipment used by measurement sites, I would expect the resonances to appear in a wide variety of test results. If human in-ear measurements are done, perhaps the natural variations would tend to smooth the results and you'd wind up back at the Harman curve in the end.

ADU's picture

Thanks again for your insights, Warren, on all of the above. I've been tied up with some other work the last few days. But I've calculated the difference between the current Harman Bass & Treble Filters Gain curve for headphones (I call it "BTFG", for short) and various room curves based on the listener preferences in Floyd's PDF, and will try to post the results tomorrow. Think you'll find it interesting.

ADU's picture

No new plots yet, but I do have a little more info to offer re the Bass & Treble Filters Gain (BTFG) correction curve shown on page 19 of this PDF...

https://dl.dropboxusercontent.com/u/16343460/AES%20139%20Presentation%20...

While making some of my comparison plots, I discovered (sort of by accident) that the BTFG curve in the above PDF is _not_ the exact room correction curve used in the new Harman Headphone Target.

I had to dig around a little bit to find the correct curve, but I'm about 99.9% sure it's the black "Preferred In-Room Loudspeaker" curve shown on slide 38 of this older 2014 PDF posted on Sean's Blog...

https://db.tt/XMgixjBP

I think there's also a slightly better graphic of the curve in his original AES "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses" paper, which dan be DL'd (for a fee) from here...

http://www.aes.org/e-lib/browse.cfm?elib=17042

(I'd post a direct link to the image, but since I don't have Sean or Harman's permission for that, you'll have to Google it.)

So why do I think this is the correct room curve, rather than the one shown in the more recent Preferred Bass & Treble Levels PDF?... Because if you subtract the dashed green curve on the following graph (which represents a loudspeaker EQ'd to a flat in-room response)...

http://cdn.innerfidelity.com/images/160503_Blog_AcousticBasisHarmanTarge...

...from the latest Harman Target curve shown in blue, the result is identical (as far as I can tell) to the black Preferred In-Room Loudspeaker curve shown in the other two links above.

I think Sean probably mentioned this in his comments above, but it seems that after doing the more recent 2015 headphone study, he and the folks at Harman essentially concluded that the previous Preferred In-Room Loudspeaker curve indicated above was correct.

The previous Preferred In-Room Loudspeaker curve is very close to the subjectively preferred room curve for all 11 listeners shown in Figure 14 on page 528 of this Floyd Toole paper...

http://www.aes.org/e-lib/download.cfm/17839.pdf?ID=17839
http://www.aes.org/e-lib/browse.cfm?elib=17839

So I'm guessin these results were from the same, or a similar study.

The Preferred In-Room Loudspeaker curve does have a somewhat smoother rise in the bass than the more recent "BTFG" curve (which is an improvememt imo). But it doesn't really have any roll-off in the upper treble. The curve is basically a flat diagonal line between 200 Hz and 20 kHz.

So... I'll go ahead and plot some difference curves based on this new/old Loudspeaker curve to see how it stacks up against some other room curves, just for grins. (I don't see much point in pursuing the newer "BTFG" curve any further, since that's obviously not what Harman is using for their latest headphone target).

I think it's pretty obvious what they're aiming for though, based on all the above graphs. The latest Harman headphone target is basically the dashed green in-ear loudspeaker curve shown on all of the various Harman Headphone Target plots, plus the black Preferred In-Room Loudspeaker curve referenced in the links above.

Sean's 2014 Blog with the link to the PDF referenced above...

http://seanolive.blogspot.com/2014/01/the-perception-and-measurement-of....

wktenbrook's picture

ADU, I think you're getting to the heart of why I wrote my article. If I can summarize a few of the curves you've cited:

  • "BTFG" curve (October 29-Nov. 1 2015. p. 19) - derived by listeners using 'method of adjustment' of bass/treble.
  • "Preferred In-Room Loudspeaker" curve (January 5 2014, p. 38) - derived by listeners using method of adjustment of bass/treble.
  • "Subjectively preferred room curve" (Toole, 2015 July/August, Fig. 14) - The lighter lines are for listeners with various levels of training from work by Olive, again using bass and treble method of adjustment.
  • My whole focus writing the article is to try to draw the curtain back and see what the reference loudspeaker-room setup measurement actually looked like on the HATS. I'm less interested in the listener preference studies listed above. Examples of in-room curves that eliminate the listening tests are:

  • Toole, 2015, Figure 13(a through c) - measured and predicted responses for speakers in various rooms - no listening tests or treble/bass adjustments.
  • Toole, 2015, Fig. 14 - dark line and shaded area are reproduced from Fig. 13(a) - no listener testing or adjustments.
  • I note that the loudspeaker curves in Toole Fig. 13(b) tend to have the 1 dB/octave diagonal line between 100 hz - 20 kHz. Even with the "six seat average" measurements cited by Toole, the lower bass is somewhat unpredictable, but the measured results are consistent with this figure used by Olive with my trend line added:

    http://cdn.innerfidelity.com/images/160503_Blog_AcousticBasisHarmanTarge...

    I then backed out this 1 dB/octave response from the 'Green Curve' to derive my curve, assuming properly managed bass will exhibit the same 1 dB/octave rise to 20 Hz (see JBL Synthesis curve).

    This presents a question that requires a value judgement: Should a headphone curve be derived from acoustic measurements under known initial conditions (Toole 2015 Figure 13a-c), or from listener adjustments (Toole 2015 Figure 14, using data from Olive)?

    If one desires a "neutral" headphone curve with the intention of simulating the good speaker-room experience, then I conclude the curve should be derived *exclusively* from the in-room speaker-HATS measurements. Listener preferences can confirm this measured curve, but they don't define it.

    I think we should stay tuned for Tyll's measurements and forthcoming papers from Olive and Welti to gain more information about stereo and multichannel measurements using the HATS.

    detlev24's picture

    I am short in time right now but I think you will find some answers in the following document:

    Listening Conditions and Reproduction Arrangements for Multichannel Stereophony
    http://www.tonmeister.de/foren/surround/texte/SSF_01_1_E_2002_v2a.PDF

    Btw, home cinema follows quite the same EQ as stereo sound reproduction, e.g., www.minidsp.com

    Jim Tavegia's picture

    My own main room has a gentle downward slope to the HF, but is very linear and it is what I have become accustomed to believe is accurate, at least for my hearing. great dialogue and thanks for sharing.

    I am waiting for Harman to put out their own "cans" from this work, but I am sure that the folks at AKG are benefiting from it. I also have two pair of affordable JBL bookshelf speakers that I find as great values and really enjoy the sound and presentation. One need not spend crazy money to enjoy very good sound.

    mat's picture

    I love everything about this article. This is a great summary with a lot of excellent points. I generally agree with just about everything said. Thanks for sharing, Tyll! And thank you Warren for writing it up!

    I also found it particularly interesting that Warren seems to speculatively support one sentiment I've had towards the current Harmen target Response curve, which seems a little too boosted around 100 Hz to me. I've never really been too keen on the shelf filter shape in the bass region and find I prefer a more consistent, gradual slope.

    castleofargh's picture

    KUDOS for the intriguing discussion, but most of all, for being able to express things in a way most people can readily understand.
    I knew that about doc Olive and his great pedagogy already, but it was nice to also see clear ideas on the other side.

    about headphone's ideal curve, I feel like we've already narrowed it down to something precise(at least compared to what headphone precision and headphone measurements are achieving). even going from measurements to preferred signature, we're still in about the same ballpark(more like a tilt than a massive change), it's very encouraging.
    the bass matter, well real bass involves more than hearing and we can't shut off other senses when interpreting sounds, so there is no point expecting that an objective frequency based approach will ever result in us feeling the same "music". on headphones, something is missing, to some adding more bass alleviates that feeling of missing something, to others, it just makes the sub too loud. I expect that part to stay at a subjective level forever and show variations.
    the same way having a proper channel mixing would also improve things for sure. listening with headphones to albums mastered on speakers, that's still the elephant in the room IMO.

    any advice for a rather cheap coupler going in the ear to try and measure my own "almost at the eardrum response"? I have enough opinions for a lifetime, but I feel that testing an actual response would be the best way to understand more of what's happening(and I don't have VIP entries at Harman like someone ^_^, also I would need a visa...).

    wktenbrook's picture

    castleofargh - I agree we can't truly replicate the bass qualities of speaker listening via headphones. On top of that, we also localize the same sound very differently during speaker playback vs. headphones, speakers cross-talk to each ear, while headphone channels are fully separated. I could go on but you get the idea.

    But despite all these differences between speakers and headphones, Harman's test listeners preferred a curve that was very close to the IEC head measurements of good speakers in a good room. That's a big takeaway.

    MRC01's picture

    My listening room measures within 4 dB of flat at the listener position from 30 Hz to 20 kHz. It does not have the 1 dB / Octave slope mentioned here. This works well for the small ensemble acoustic music that makes up most of my listening. This article may explain why this response may be subjectively perceived as "lean" or "bright".

    The music recordings used for the test do not have inherently flat frequency response. How much does the way music is recorded affect the results? Would people want the room or their seating arranged to yield same 1 dB / Octave slope when listening to live acoustic sounds?

    Are people so far removed from live acoustic music they forget what it sounds like and strive for euphonics?

    wktenbrook's picture

    A flat steady state room measurement at the listening position in a typical room will be too bright, including good rooms with properly controlled reverberation time. The only speaker measurements that should be flat are near the speaker axis in an anechoic chamber, or the first arrival sounds from the speaker in the listening room before any early reflections arrive. If you measure steady state sound at a listening seat, either sine waves or calibrated noise spectrum, you will be including all sound in the room, both direct and reflected, and that spectrum should have the 1 dB/octave slope, or something close to it.

    Also, check out ADU's excellent posts in this thread calling attention to the need for a holistic approach to the equalization question. The only way the simple 1dB/octave slope can really work correctly is if the speakers are anechoic flat on axis with smooth directivity, and the room is set up to yield controlled reverberation time across the audio spectrum. If the speakers or room are not up to the task, equalization can't correct the problems.

    Bringing this back to headphone curves, Harman has speakers that can attain these goals (JBL Professional and Revel) and they have the Harman Reference Room specifically designed to meet reverberation time specs. This puts Harman in position to measure the listening seat with a head and torso simulator and record what a reference speaker-room set-up sounds like. That's the essence of the Harman headphone target curve.

    You mention that music recordings do not have inherently flat response. I agree that music is recorded with all kinds of crazy approaches over the years and it definitely affects the results. Dr. Floyd Toole complains this is one aspect of the "circle of confusion." But we need to take a stand for technical standards that everyone can understand, use, and benefit from to get better sound. If you ask Innerfidelity contributor Bob Katz, he would likely reply his mastering studio set-up is similar to the Harman reference arrangement used to measure the headphone target curve.

    Finally, the acoustics of live concert venues is a fascinating topic (look up 'shoebox concert hall'). However, listening room acoustics are completely different. A concert hall doesn't follow the 1 dB/octave curve. Acousticians and architects have their own set of rules for larger spaces and genuine instruments.

    MRC01's picture

    This inspired me go back and re-measure my listening room. It is indeed +/- 4 dB at the listener position; but the -4 is at 20 kHz and +4 is at 80 Hz (smoothly rolling off to -4 dB @ 30 Hz). It's similar to the recommended -1 dB / octave slope - except for a small bump centered at 1 kHz (+3 dB @ 1 kHz, 3 dB / octave slope). This bump is likely the source of sounding a bit on the lean side.
    It took careful arrangement, huge tube traps and copious use of thick acoustic foam to get the response as good as it is. I don't know what kind of room treatment can eliminate that little bump. Perhaps only a digital parametric equalizer will do the trick.
    Thanks for a great article that improved my understanding of this interesting topic.

    Kjetil K's picture

    Thanks for a really good WEB site!

    There is one thing I miss in the discussion here, and that is the fact that when listening to loudspeakers, one ear receives sound from both left and right speaker. That actually gives a "changed" frequency response because of out of phase sounds.
    For a setup of speakers at +-30 degrees, this typically will give a loudspeaker system a cancellation around 1,7 KHz, and amplification around 3,2 KHz, and so on. This of course changes at different speaker angles.
    This does not happen when listening to headphones, as left channel is only received in left ear.
    So what happen when the sound engineer is recording?
    Typically sound sources "in front" will "miss" energy in the 1,6 KHz region, so he turns this up, and turns down a bit the 3 KHz region. For sound source more to the left or right, there is less change in perceived frequency response, so with these sources the engineer does less EQ-ing.
    How will this then sound in a headphone, even if the headphone is correctly EQ'ed to the Harman curve, or some other curve?
    Probably the sound coming from the front in the mix will sound sharp as there will be too much energy in the 1,6KHz region. But it all depends on the engineer... If he/she didn't EQ this source in the first place, the music will probably be very laid-back when listening on speakers, and good/perfect on headphones... :)

    /Kjetil K

    MRC01's picture

    The figures you quote are interesting, because my speakers measure at the listening position about +2 dB @ 1.7 kHz and -2 dB @ 3.2 kHz. Now you have me wondering if this was by design.

    If live acoustic music were being played in the room, the ear would still have a dip in response around 1.7 kHz and a peak around 3.2 kHz, no? So reproducing these frequencies FLAT would be essential to preserving the natural sound.

    In recording, it would depend on how the event was miced. If individually close miced, these dips & peaks would not be in the recording, so you'd want the room to reproduce them on playback - flat response. But if the recording was miced from a distance (like a Blumlein pair of matched wide range mics), the recording would capture these peaks and dips, just like human ears would, so flat response on playback would double the effect, making it sound less realistic.

    Kjetil K's picture

    It might be that the the system was designed to give a little boost at 1,7kHz and a little lower at 3,2kHz, but I think +-2dB can also be just a coincidence. But I know that many loudspeaker manufacturers actually select the 3kHz as crossover frequency just for this reason, to get a dip in power response at 3kHz.
    To the second question: A real source e.g a singer in front of you will not give this frequency response change. This is something that happens when two loudspeakers reproduce the same sound to give perception of a single source in front of you.
    So this effect is something stereo on loudspeaker introduces, and dependent on what the recording engineer uses, headphones or loudspeakers when mixing, the result will either be good for loudspeakers or good for headphones, or something in between.

    MRC01's picture

    Ah, that makes sense - point source sounds can't have this effect. So natural sounds that are not point source, could - though this would seem rare, if possible.
    The speakers in question are Magnepan 3.6/R; the ribbon tweeter is about 5' long vertically oriented and is crossed over from the midrange around 1.5 kHz.

    RAJIB123's picture

    Have you heard all the talk about “web site design”? Do a quick Google search to see why people from all over the world can’t say enough about this design that helps you make money online.
    https://diffrentstypeofinformation.blogspot.co.uk/2016/05/html-tutorial-...
    No much talks here. Simply put "web site design" works! It worked for me and it can work for you as well. You can use it as a part time job or even do it full time. To learn more, please do a quick Google search for it.
    https://youtu.be/L19bO0Pzia8
    https://youtu.be/TmSzXt7H28I
    https://youtu.be/fit3NLY7mJo

    ADU's picture

    Please see my new post re the Bass & Treble Filters Gain curve on page 1 of the Comments. I think the info there may shed a little better light on exactly how the latest Harman Target was derived.

    (Hopefully all the links are working correctly.)

    Bob Katz's picture

    Hello Warren, Hello Sean, Hello Tyll! This is excellent work. I feel drawn to this discussion (wonder why) :-). Of course until mastering rooms are standardized (I doubt they will ever be), the circle of confusion will prevail. There is some degree of truth to the current situation, at least in popular music: mixing engineers frequently work with near-field monitors and then mixes are frequently mastered by mastering engineers in well-aligned midfield rooms. Classical music is often the exception to that, and frequently, classical music is not mastered or post-processed after mixing.... instead it is "mixed and mastered" at once in the reference room used by the classical mixing engineer.

    ----

    My mastering room is extremely calibrated and I have total control over its response. I've arrived at its response using a scientific method but also (by necessity) empirically. In other words, there is a necessary deviation from flat, but how much deviation to use was arrived at through subjective and practical means.

    Warren has argued that the ideal loudspeaker/room response should have a continuous 1 dB/octave tilt from bottom to top, but I have found that a flat response from bass through to 1 kHz, followed by a rolloff that is about 6 dB down at 20 kHz, works very well for me. And if mastering rooms ever become standardized, I would tend to argue in favor of that because historical precedent must prevail, and the vast majority of my reference recordings perform very well in a room with this response.

    Note that I measure my response using Acourate, which uses a variable window that begins circa 200 ms wide in the bass region and gradually becomes near anechoic at the high frequency region, based on the authority Jim Johnston, AES and IEEE fellow. You can read his papers via AES preprints 7263, 8314 and 8379. Ironically, I have not been able to tolerate a flat response using this method, so I still have to have an HF rolloff, so either my measurement is not sufficiently anechoic at high frequencies, or there is something missing in my understanding. Thus I have to be empirical as well as scientific. I arrived at the high frequency rolloff (target) in my room by listening to about 50 of the best recordings that I know as references. The brightest of them do not sound too bright and the dullest do not sound too dull, with the inference that neither will my masters. Chicken and the egg, granted, but at least it's a method that produces very satisfied clients and listeners, or I wouldn't be a successful mastering engineer. If this method preserves the status quo in recordings, I think it's still pretty good.

    -----

    I want to point out that every tenth of a dB of bass response change is also perceived as a treble response change. In other words, slightly more HF rolloff is also perceived as an increase in bass. This also occurs when mastering, if I make a master that's a little brighter, it also seems a little thinner. None of this is new information, but it's important to realize that my room which is flat in the bass and is 6 dB down at 20k, could sound subjectively similar to a room which has a "Tenbrook tilt".

    For the past year or so, I've been using a wide Q filter centered at 20 Hz in the target response. It's currently + 0.5 db at 20 Hz because the number of listeners in the room affects the bass response. Yes, our bodies are big bags of water, so I've found about a half dB (subjectively determined, but also measured) bass loss when a second person joins me in the mastering room, so to make an average that works well, by necessity my room is +0.5 at 20 Hz, rolling off smoothly so that by 110 Hz it's measurably flat, with one person in the room! It's part of the empirical method I have to employ, and it works very well, I can accomodate to that very easily. I could also alter the target response depending on the number of people in the room, but I haven't gotten around to that.

    That's the news from Lake Woebegone. I await seeing Tyll's measurements of the Harman room with alacrity!

    wktenbrook's picture

    "...I have found that a flat response from bass through to 1 kHz, followed by a rolloff that is about 6 dB down at 20 kHz, works very well for me. And if mastering rooms ever become standardized, I would tend to argue in favor of that because historical precedent must prevail, and the vast majority of my reference recordings perform very well in a room with this response."

    Bob, thanks for your comments.

    Sounds like you've arrived at a curve that works well and is reasonably close to the Bruel & Kjaer 1974 study, which certainly has 'historical precedent'. My June 28 post suggests a head simulator measurement in your room would look something like an Audeze LCD series frequency response. That's unsurprising since I recall one of your reference headphones is the LCD-X.

    "...Warren has argued that the ideal loudspeaker/room response should have a continuous 1 dB/octave tilt from bottom to top..."

    I don't want to give the wrong impression. I am seeking the listening position curve that would naturally come from a great pair of speakers in a great-sounding room - but that's a lot of variables to conquer and I'm open to different possibilities. As I reviewed in-room data from speakers designed to be flat in anechoic chamber and having good bass extension, like the big JBL Professional or Revel models, it appeared that the 1 dB/octave line was a natural in-room behavior of that category of loudspeaker - and it was a simple starting point. Whether that slope correctly reflects the generic behavior of anechoic flat speakers and is good for mastering is up for debate.

    Bob Katz's picture

    Hola Mr. wktenbrook!

    Years ago Peter Walker of Quad came up with what is called the "tilt" EQ. It's centered at nominally 1 kHz and subtly raises the bass range as it lowers the treble, etc. It's clear that the ear reacts to low and high frequency response in a kind of yin and yang, that is, more bass is often perceived as less treble. In mastering, that's one of the reasons a smile curve is so common :-). In mastering I try to guard against creating a smile by being aware that as I add some treble, if I then think the bass is thin, instead of adding more bass, I just take back some of the treble I had boosted because obviously we had reached diminishing returns.

    The bottom line of this psychoacoustic phenomenon is that there is more than one way to skin a headphone response and both ways may come out subjectively the same or at least very similar. So whether this turns out to be a continuous tilt from bass to treble or (as I prefer) flat bass with an HF rolloff, the net sonic result can very easily turn out virtually indistinguishable. Especially depending on the center of your turnover or the center of the tilt frequency. Another approach is to be flat from say, 100 to 1k, have a slight rise if necessary below 100 and a slight rolloff above 1k. Whether in that case the bass rise should be a tilt or a shelf is debatable, but in that case my experience that a subtle linear tilt from 100 down to 20 sounds quite good. All of these results have been tested by me in a listening room with a calibrated measurement microphone and using a particular FFT variable window width. The number of variables at even measuring what we call "flat" are so large that EQ'ing headphones or loudspeakers will continue to be both a science and an art for many years to come. But we all agree in the general shape: The devil is in the details.

    By the way, while the LCD-X is one of my favorite cans, it's not entirely satisfactory out of the box. I routinely add some top end to try to get some openness. My one-time impression of the LCD-4s at an AES show was that the LCD-4 is much closer to sounding right without needing any EQ at all.

    dbmcclain's picture

    Hi,

    I keep seeing this term bandied about, especially in regard to Harman employee test subjects. Can you define the term "Trained Listener'?

    Love this discussion!

    - DM

    donjoe's picture

    Speaking of LCD response curves, what if listeners have different preferences - especially in terms of bass quantity - depending on the membrane speed? Has anyone considered evaluating listener preferences with "slower" dynamic headphones vs. planars?

    X