Head-Fi.org › Forums › Equipment Forums › Sound Science › AES 2012 paper: "Relationship between Perception and Measurement of Headphone Sound Quality"
New Posts  All Forums:Forum Nav:

AES 2012 paper: "Relationship between Perception and Measurement of Headphone Sound Quality" - Page 3

post #31 of 130

Just read the paper 8740 thoroughly, with the help of a good friend. I can already tell HP1 is LCD2 v2, and HP2 is AKG K701. smily_headphones1.gif Sorry, I've done too many measurements! LOL

 

 

HP1: Audeze LCD2 v2

HP2: AKG K701

HP3: Bose QC15

HP4: AKG K550

HP5: Monster Beats LE

HP6: Crosfade LP

 

Dr.Olive's point is that the flat diffuse-field reference is not an ideal target for headphones, due to much deviation from subjectively assessed results. It is important to note that K701 and LCD2 v2 differ in 'headphonic' weight-class, considering how SUPER-linear LCD2 v2 behaves from the sub-bass to 10 kHz (with a slight nick, roughly -6 dB @ 4 kHz), not to mention that K701 loses a sub-bass response even with a slight acoustic leak due to its higher acoustic output impedance. LCD2 v2's sweet spot is relatively larger too, radiating the sound wave more uniformly. In addition, LCD2 v2 is one of the mostly diffuse-field response oriented headphones, if not better than K701. 

 

Quite interestingly, Dr.Olive's assessment with the diffuse-field reference directly contradicts the finding from the paper 8744, which was presented at the same convention as well. The german Fraunhofer institute tells a totally different story, claiming the ISO diffuse field reference is, in terms of listener preference, on par with, or better than three different loudspeaker-matched references calibrated by 3 different expert listeners in a ITU-R BS1116-1 condition. Not to mention that Lorho's reference turns out to be a complete bogus! It is very fortunate I happen to use the same ISO target too.

 

BTW it is nice to see some blocked-meatus measurement results as well. As stated by Møller and Hammershøi, although interpersonal deviations may vary among different headphone types, there is definitely a chance for general agreement, much less deviating than open-meatus measurements. 

 

Also it is very interesting to see LCD2 v2, which has an unexaggerated wide sub-bass response, was rated the flattest among other bass-boosted headphones without any type of tactile stimuli introduced. Is it possible that, as I have mentioned many times, a flatter & wider bass is more preferable than a simple boosted bass?


Edited by udauda - 11/6/12 at 5:55am
post #32 of 130
It's all Greek to me!
post #33 of 130
Quote:
Originally Posted by udauda View Post

Just read the paper 8740 thoroughly, with the help of a good friend. I can already tell HP1 is LCD2 v2, and HP2 is AKG K701. smily_headphones1.gif Sorry, I've done too many measurements! LOL

 

 

HP1: Audeze LCD2 v2

HP2: AKG K701

HP3: Bose QC15

HP4: AKG K550

HP5: Monster Beats LE

HP6: Crosfade LP

 

Dr.Olive's point is that the flat diffuse-field reference is not an ideal target for headphones, due to much deviation from subjectively assessed results. It is important to note that K701 and LCD2 v2 differ in 'headphonic' weight-class, considering how SUPER-linear LCD2 v2 behaves from the sub-bass to 10 kHz (with a slight nick, roughly -6 dB @ 4 kHz), not to mention that K701 loses a sub-bass response even with a slight acoustic leak due to its higher acoustic output impedance. LCD2 v2's sweet spot is relatively larger too, radiating the sound wave more uniformly. In addition, LCD2 v2 is one of the mostly diffuse-field response oriented headphones, if not better than K701. 

 

Quite interestingly, Dr.Olive's assessment with the diffuse-field reference directly contradicts the finding from the paper 8744, which was presented at the same convention as well. The german Fraunhofer institute tells a totally different story, claiming the ISO diffuse field reference is, in terms of listener preference, on par with, or better than three different loudspeaker-matched references calibrated by 3 different expert listeners in a ITU-R BS1116-1 condition. Not to mention that Lorho's reference turns out to be a complete bogus! It is very fortunate I happen to use the same ISO target too.

 

BTW it is nice to see some blocked-meatus measurement results as well. As stated by Møller and Hammershøi, although interpersonal deviations may vary among different headphone types, there is definitely a chance for general agreement, much less deviating than open-meatus measurements. 

 

Also it is very interesting to see LCD2 v2, which has an unexaggerated wide sub-bass response, was rated the flattest among other bass-boosted headphones without any type of tactile stimuli introduced. Is it possible that, as I have mentioned many times, a flatter & wider bass is more preferable than a simple boosted bass?

 

 

 

Thanks for the thorough review of our paper, which I hope to soon discuss in my blog. I would disagree with your statement that the Audeze model we tested is closest to the diffuse field (DF) response recommended by the IEC spec. When measured on our GRAS 43 AG coupler/ear simulator the Audeze had the "flattest response" than any headphone we tested. If it had a true DF response it would exhibit the typical 10 dB peak at 3kHz which it lacked when measured on the ear simulator. The K701 actually more closely approximates the DF response.

 

The other AES paper you refer to  by Fraunhofer presented at AES133rd was  "Identification and Evaluation of Target Curves for Headphones" where they compared two or three headphones equalized to different targets including DF and a modified DF (Lorho 2009); the latter reduces the DF  3kHz peak measured at the ear drum from 10 dB to ~3 dB. These two targets were compared  to  the original unequalized headphone, and 3 target curves determined by "experts" listening to a 5.1  loudspeaker room in an ITU- BS116 listening room. No  acoustic measurements were offered in the paper or by the author at the presentation regarding what those expert targets were, nor did he provide anechoic and in-room measurements of the loudspeakers/room used by the "experts"  a reference from which they derived their preferred target responses. Based on MUSHRA-type  listening tests, the DF curve was generally less preferred to the expert target curves. In cases where the DF was highly rated this could be explained because the original unequalized phone had a terrible frequency response, and any sort of equalization made it sound better. Not exactly proof that the DF calibration is preferred. And it's too bad that not enough information was provided in the paper to know what the "expert" target curves were.

 

I believe that there are better headphone response targets than the DF calibration based on the fact that a)  we don't typically listen to stereo recordings in diffuse fields and b) the recent headphone papers reporting listening test results suggest DF calibration is sub-optimal in terms of sound quality.  What the optimal target should be is a topic of continuing research by our research group, and others. Stay tuned.


Edited by Tonmeister2008 - 11/10/12 at 6:50pm
post #34 of 130

Thank for sharing your insight with us, Dr.Olive!

 

Quote:

I would disagree with your statement that the Audeze model we tested is closest to the diffuse field (DF) response recommended by the IEC spec. When measured on our GRAS 43 AG coupler/ear simulator the Audeze had the "flattest response" than any headphone we tested. If it had a true DF response it would exhibit the typical 10 dB peak at 3kHz which it lacked when measured on the ear simulator. The K701 actually more closely approximates the DF response.

The mid-frequency peak is definitely there, Try it on a HATS, especially on B&K 4128c. LCD2 v2 is very flat with a mid-high frequency boost, resembling a standard diffuse-field reference, but somewhat downscaled in degrees, which gives it roughly a -5dB/oct tilt in the mid-high frequency range upon calibration.

 

 

700

 

This is LCD2 measured on Neumann KU100, which is diffuse-field referenced, published by AUDEZE themselves. (According to Martin Schneider of Neumann, "Both KU81 and KU100 are diffuse-field equalized, meaning for diffuse reverberant sound the sonic impression should be neutral.") You can definitely see the downward tilt I mentioned. Tyll Hertsen's LCD2 measurements are in good agreement with AUDEZE's calibration chart as well.

 

 

K701 is DF equalized too with a -3dB/oct tilt, but it lacks good amount of sub-bass below 40 Hz, not to mention its high frequency response is very uneven with more peaks and dips.

 

No  acoustic measurements were offered in the paper or by the author at the presentation regarding what those expert targets were, nor did he provide anechoic and in-room measurements of the loudspeakers/room used by the "experts"  a reference from which they derived their preferred target responses. Based on MUSHRA-type  listening tests, the DF curve was generally less preferred to the expert target curves. 

I totally agree on your point that they did not provide enough information. Wonder why someone from Fraunhofer institude would miss that..? Sent the author an email asking the reason why certain data have been omitted, I hope to get his response soon. Dr.Fleischmann, the author of the paper 8744, sent me an email back, promising to send me the answer within next few days. 


In cases where the DF was highly rated this could be explained because the original unequalized phone had a terrible frequency response, and any sort of equalization made it sound better. Not exactly proof that the DF calibration is preferred. And it's too bad that not enough information was provided in the paper to know what the "expert" target curves were.

They used not only Sennheiser PX200, but also STAX SR-404, Sennheiser HD600.

 

I believe that there are better headphone response targets than the DF calibration based on the fact that a)  we don't typically listen to stereo recordings in diffuse fields and b) the recent headphone papers reporting listening test results suggest DF calibration is sub-optimal in terms of sound quality.  What the optimal target should be is a topic of continuing research by our research group, and others. Stay tuned.

If your current speculation happens to be correct, then Etymotic Research ER-2 or Monoprice 8323, which has one of "the flattest uncalibrated response" I've ever seen, should be the most preferred headphones.

 

a) Headphones have diffuse-field oriented localization characteristics, according to Hirahara and Theile. While staying close to the conventional principle of DF, giving a slight downward-tilt will yield much better preference, just like how Audeze tuned their LCD2 v2. (and less peaks and dips of course) 

 

b) Since AES studies dealing with same subject came up with two different conclusions, I think I will defer my decision until more conclusive results come up. Will stay tuned for sure.


Edited by udauda - 11/13/12 at 1:23am
post #35 of 130

popcorn.gif
 

post #36 of 130

If any of you are members of Linkedin you can contact Sean via his link

http://www.linkedin.com/in/seanolive

 

 

Sean Olive

President-Elect at Audio Engineering Society, Director Acoustic Research, Harman International

 

Director of Acoustic Research

Specialties

Expertise in scientific research related to perception and measurement of sound quality in audio recording and reproduction systems:microphones, loudspeakers, amplifiers, listening rooms, automotive audio systems, and binaural and spatial audio algorithms. I direct the Corporate R&D Acoustics Group and am responsible for sound quality benchmarking of Harman's consumer, professional and automotive audio systems.

post #37 of 130
Thread Starter 

Finally got a chance to read the paper online at http://www.aes.org/e-lib/browse.cfm?elib=16446  This paper is a good step towards more subjective testing for headphones, which I think is really the right way to do science in this field.

 

My one comment would be that I wish the testing could have taken the "virtual" route as in cited work [9]. It appears that the authors' hypothesis had already been that frequency response (spectral balance) would be the most important factor in sound quality, and the paper findings confirm it. If that's the case, why not isolate this one variable using equalization, leaving behind all the other variables that differ among headphones? I don't think we need to go as far as binaurally recording each headphone as suggested in Section 3.2.3, but simply to stick with a single high quality headphone, while testing different frequency responses using DSP. That would basically combine the Fraunhofer's paper's approach with the subjective evaluation and objective measurements in this paper.

 

Thanks again to Tonmeister for the research. I hope these papers will advance the headphone field the same way Floyd Toole's work did for loudspeakers. As an owner of a loudspeaker that embodies that science, I am reminded every day how much further headphones have to go to catch up.

post #38 of 130

Finally! A well conducted subjective study on musical perception!

 

Hopefully this means that companies will jump on a "linear" bandwagon and we all have psudo studio monitors in a few years. 

 

Gosh i can only wish for flat frequency response coaxials for cars..that would be the best thing to happen to car audio ever.

post #39 of 130

popcorn.gif

post #40 of 130

At last, I received the answer from Dr.Fleischmann!

 

First, on the claim that the flat eardrum(DRP) response being the most preferred target:

..From a physical point of view, we started of the design goal, that a headphone should more or less compensate for the missing transfer path between loudspeaker and listener's ear. [in the listening test, this design goal was rated quite high] Basicallly, this design goal [now described for DRP] is caused by a superposition of
 
1.) sound pressure doubling (+6dB) because of the wavelength of radiated sound by an external sound source is getting to the dimensions of our head [above ~2-3 kHz], can e.g. be calculated assuming our head is a sphere.
 
2.) the ear-canal acting seen a pipe with length approx 2-3cm, diameter ~0.8cm which is rigid at one end and thereby acting like a quarter wave resonator, also having the first peak at the same freq. like
 
3.) any other influences like pinna, shoulder etc.
 
So i don't see the point why a flat response at DRP should make sense..
I see his point on this absolutely valid; can't simply disregard the psychoacoustic aspects of all those missing HRTFs. Now all I wonder is why Dr.Olive's LCD-2 v2's data were so off compared to other simulator (DRP or not) LCD2 measurements. 

 

Second, on the reference condition at Fraunhofer Institute:

The room is conform to ITU-R BS 1116. There is a complete AES paper...about this room and the charcteristics. The loudspeakers used are 'Geithain RL901K', they can also be seen in the paper..

And I took a snapshot out of the printed version of the paper. Pretty darn good.

1000

 

Third, on his 'expert curves' disclosure:

..i am sorry that i can't show these. what i can tell, and this can also be seen in the listening test, that expert 2 and 3 showed mainly similar behavior like the ISO curve..
It is quite unfortunate he can't disclose the data, possibly due to something like their own NDA. (Or they are coming up a patent with this technique?) 

Edited by udauda - 11/15/12 at 6:50pm
post #41 of 130
Quote:
Originally Posted by udauda View Post

At last, I received the answer from Dr.Fleischmann!

 

First, on the claim that the flat eardrum(DRP) response being the most preferred target:

 

 

..From a physical point of view, we started of the design goal, that a headphone should more or less compensate for the missing transfer path between loudspeaker and listener's ear. [in the listening test, this design goal was rated quite high] Basicallly, this design goal [now described for DRP] is caused by a superposition of
 
1.) sound pressure doubling (+6dB) because of the wavelength of radiated sound by an external sound source is getting to the dimensions of our head [above ~2-3 kHz], can e.g. be calculated assuming our head is a sphere.
 
2.) the ear-canal acting seen a pipe with length approx 2-3cm, diameter ~0.8cm which is rigid at one end and thereby acting like a quarter wave resonator, also having the first peak at the same freq. like
 
3.) any other influences like pinna, shoulder etc.
 
So i don't see the point why a flat response at DRP should make sense..
I see his point on this absolutely valid; can't simply disregard the psychoacoustic aspects of all those missing HRTFs. Now all I wonder is why Dr.Olive's LCD-2 v2's data were so off compared to other simulator (DRP or not) LCD2 measurements. 
 

It may depend on the recording... How about binaural ones...

 

And... yeah popcorn.gif


Edited by ultrabike - 11/15/12 at 10:26pm
post #42 of 130
Quote:
Originally Posted by ultrabike View Post

It may depend on the recording... How about binaural ones...

 

IMHO, off topic. Regardless, here's a good read for that. A flat eardrum (DRP) response should work only when recordings contain complete HRTF-related information measured at the DRP. Other than that, it is not applicable due to the facts that:

 

1. Almost all of conventional binaural recordings, such as Neumann's dummy head recordings, are pre-equalized.

2. A flat blocked-meatus (EEP) response, which is obtained via inverse filtering, still has non-directional ear canal resonance when measured at the eardrum. (including Harman Kardon's BRS technique)

 

There's a reason why Etymotic Research recommends their ER-2 for hearing related researches only.

post #43 of 130

Can I ask a couple of questions?  I am the second author on the "Relationship..." paper.  I apologize for coming in a bit late to this discussion.  A couple of questions:

 

Udauda, your plot of the Audeze confirms our assertion that the Audeze does not have a "DF" response, otherwise it would be flat on your DF equalized HATS, no?  It is clearly missing all or most of the 3 kHz resonance.  It is true that there is a broad upwards tilt when measured at DRP with no correction (I prefer to stick to uncorrected measurements for clarity), but that is not a DF response.  Or did I miss something.

 

Also:

 

Quote:
I believe that there are better headphone response targets than the DF calibration based on the fact that a)  we don't typically listen to stereo recordings in diffuse fields and b) the recent headphone papers reporting listening test results suggest DF calibration is sub-optimal in terms of sound quality.  What the optimal target should be is a topic of continuing research by our research group, and others. Stay tuned.

You replied:

Quote:
If your current speculation happens to be correct, then Etymotic Research ER-2 or Monoprice 8323, which has one of "the flattest uncalibrated response" I've ever seen, should be the most preferred headphones.

 

Sean just said that there are better responses than the "DF" response.  We don't know what they are.  Sean did not say that a flat response, like the ER-2, would be best.  I'm not sure how you got there.

 
Otherwise I agree with most of what has been said on this thread.  
post #44 of 130

Can I ask a couple of questions?  I am the second author on the "Relationship..." paper.  I apologize for coming in a bit late to this discussion.  A couple of questions:
Of course, as I am thrilled to have you, Mr.Welti at Harman Kardon, joining with us. I am sure others are too.

 

 

 

Udauda, your plot of the Audeze confirms our assertion that the Audeze does not have a "DF" response, otherwise it would be flat on your DF equalized HATS, no?  It is clearly missing all or most of the 3 kHz resonance.  It is true that there is a broad upwards tilt when measured at DRP with no correction (I prefer to stick to uncorrected measurements for clarity), but that is not a DF response.  Or did I miss something.

The plot, which is measured with diffuse-field compensated Neumann KU100, is equivalent to a DRP measurement with the ISO diffuse-field reference response invert-equalized. Thus AUDEZE LCD-2 v2 IS indeed diffuse-field equalized, yet the degree is slightly less with a -5 dB per octave downward tilt, as LCD-2 v2's DRP data (B&K 4128C & Neumann KU100) indicate there's only ~10 dB peak @ 3 kHz. There's nothing missed on my end.

 

 


Sean just said that there are better responses than the "DF" response.  We don't know what they are.  Sean did not say that a flat response, like the ER-2, would be best.  I'm not sure how you got there. Otherwise I agree with most of what has been said on this thread.  

I am very please to know Mr.Welti agrees with me. And this is how I got there. Dr.Olive clearly states, 

..The most preferred headphones generally had the smoothest and flattest measured amplitude response when measured on GRAS 43AG ear simulator..The primary difference in amplitude response between the most preferred headphone (HP1) and next runner up (HP2) was the lack of the 12 dB peak centered between 2-3 kHz that is required to meet the IEC DF equalization. In terms of sound quality, HP1 was judged to have a neutral spectral balance.. p.12

Since GRAS 43AG is DRP oriented, "the smoothest and flattest amplitude response" must mean a flat response measured at the eardrum. As he measured LCD-2 v2 as "flat" with 43AG, corresponding data must be "DRP flat" as well. So I was merely listing a few of such flat DRP-measured headphones, so that we can have better understanding in what those products are. 

 

Yes, we do not know what they are just yet, but for now, IMHO, as Dr.Fleischmann has stated, the "better response" should be very close to the "DF response" unlike Dr.Olive and you claimed on the paper 8744. More will be revealed, as more subjective evaluations of headphones come up in the upcoming years.

post #45 of 130

Thanks for welcoming me to this forum!

 

 

Quote:
The plot, which is measured with diffuse-field compensated Neumann KU100, is equivalent to a DRP measurement with the ISO diffuse-field reference response invert-equalized. Thus AUDEZE LCD-2 v2 IS indeed diffuse-field equalized, yet the degree is slightly less with a -5 dB per octave downward tilt, as LCD-2 v2's DRP data (B&K 4128C & Neumann KU100) indicate there's only ~10 dB peak @ 3 kHz. There's nothing missed on my end.

 

I will restate my opinion one more time and then let it go.  Your plot is a full -10 dB from 2 kHZ up, with respect to the ISO DF you reference.  For me that would be a large enough difference to say that it is not "DF" equalized, with respect to the ISO DF, or any other DF curve (there are a number of other ones and not all the same).  Your "tolerance mask" may be larger than mine smile.gif

 

As for flat being preferred, yes it turned out that in this limited study that was more or less the case, although the Audeze is not really flat (you are in fact saying it is DF).  That fact and the limited scope of our first set of tests, would not allow us to say that "flat is preferred".  We were only speaking of the six headphones we studied.  We are currently working on a new test, and so far, the "flat" (that is, equalized absolutely flat at DRP) headphone is NOT likely to be preferred over the other eq's we are investigating.  As Sean said, "Stay Tuned"!!

 

On another note, I am going to start a thread on measurement of headphones.  The subject will be "to use pinna or not to use pinna".  Not trying to cross-post, but I though I'd mention it so you can look out for it.  

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › AES 2012 paper: "Relationship between Perception and Measurement of Headphone Sound Quality"