Headphone Measurements Using Brüel & Kjær 5128 HATS

[GaryH]

That paper is quite misleading.

This paper? Although I do have some reservations about it, I wouldn't say it's misleading, it just uses a different method (cluster analysis based on headphone frequency response preference) of determining bass preference compared to the tone control method-of-adjustment of the paper you linked.

Now this I would say is misleading. When leakage was controlled, there was no difference in preferred bass level with and without loudness normalization in that in-ear headphone study:

There was a slight difference in average preferred bass shelf frequency with and without loudness normalization, but considering the error ranges this is not statistically significant:

I don't know why they're calling that the 2013 Harman target. That's the green curve below, 'Loudspeaker equalized to flat in-room response', which of course is not neutral, lacking in bass, and was just chosen as a somewhat arbitrary starting position for the MOA studies. The black curve below is the actual 2013 Harman target, as @Sean Olive explains in this post.

 

[HRTF_Enthusiast]

This paper? Although I do have some reservations about it, I wouldn't say it's misleading, it just uses a different method (cluster analysis based on headphone frequency response preference) of determining bass preference compared to the tone control method-of-adjustment of the paper you linked.


Now this I would say is misleading. When leakage was controlled, there was no difference in preferred bass level with and without loudness normalization in that in-ear headphone study:

There was a slight difference in average preferred bass shelf frequency with and without loudness normalization, but considering the error ranges this is not statistically significant:

I don't know why they're calling that the 2013 Harman target. That's the green curve below, 'Loudspeaker equalized to flat in-room response', which of course is not neutral, lacking in bass, and was just chosen as a somewhat arbitrary starting position for the MOA studies. The black curve below is the actual 2013 Harman target, as @Sean Olive explains in this post.

If they used the flat in-room response, isn't the mean bass gain level basically flat on 5128?

 

[solderdude]

imo the fidelity is audibly compromised at elevated levels even if not audibly clipping. Who knows how this could have affected the test?

That's your opinion. It looks like you don't own DT990. I do and have for over 15 years and have used it a lot in the past with that or similar EQ. And even at somewhat higher levels it sounds fine.
I'll admit I don't use it anymore as I have better headphones but the DT990 with EQ is not an audibly compromised headphones at sensible loud SPL.

 

[HRTF_Enthusiast]

That's your opinion. It looks like you don't own DT990. I do and have for over 15 years and have used it a lot in the past with that or similar EQ. And even at somewhat higher levels it sounds fine.
I'll admit I don't use it anymore as I have better headphones but the DT990 with EQ is not an audibly compromised headphones at sensible loud SPL.

I have used DT990. Admittedly, I haven't used it with EQ, but the distortion was not enjoyable.

 

[solderdude]

We have owned different DT990 then or listen at quite different levels.
For the price, and with EQ, it is an enjoyable and comfortable headphone.
You can ask Nouvraught

 

[HRTF_Enthusiast]

We have owned different DT990 then or listen at quite different levels.
For the price, and with EQ, it is an enjoyable and comfortable headphone.
You can ask Nouvraught

If it doesn't audibly distort for you, what do you mean by "better headphones"? Ones that sounds better without EQ?

 

[solderdude]

If it doesn't audibly distort for you, what do you mean by "better headphones"? Ones that sounds better without EQ?

Yes, I have a few headphones that are indeed tonally more correct without (or with minimal EQ or modifications) and/or have better 'separating' qualities or are more comfy or have different use cases or combinations of those factors.

That doesn't mean the DT990 with EQ and used at normal to somewhat louder than normal listening levels can not be used to evaluate sound quality or bass preference and it does not mean those headphones would be better suited for those tests either.

 

[solderdude]

Considering the average dBA for headphones wasn't loud, maybe most of them did not encounter distortion. However, I am still curious about the fidelity compromise the Harman shelf adds as illustrated in Amir's DT990 review.

Amir's EQ for the DT990 does not shelve low frequencies. It only pulls down 150Hz some 2dB.
My EQ for the DT990 (at 20Hz) is +1dB and pulls down the mids a good 5dB at around 120Hz (very wide band)
So.. there is no need for heroic EQ in the lows.
It could have been different if Harman used something like th AKG500/501/601/701/702 which requires quite a lot of boost in the sub lows/lows.

I think for determining bass preference at not loud levels an EQ'ed DT990 isn't bad to start with and probably a better choice than DT880 (of old) which does require quite some boost at the lowest audible range (20-40Hz).

We would need more insight on Amir's loudness levels.

I believe he listens at 'normal' levels for enjoyment. Where normal would be around 75dBA average ?
For assessment of bass levels at higher levels it seems 'shaking of his earlobes' is an indicator and read here and there he uses earplugs when listening for issues like distortion at higher SPL.

 

[HRTF_Enthusiast]

I find the data for this test to be quite interesting. Nine out of the thirty-three listeners preferred 0 dB or lower bass gain. The mean bass gain is close to 1 dB higher than the median bass gain presumably due to the two outliers. Twelve of the listeners in this test were considered trained and only eight of the total listeners had bass gains at or above 5 dB. The data could be made a lot more useful if we could see which listener IDs were trained .

The cluster analysis paper @GaryH linked used data from https://www.aes.org/e-lib/browse.cfm?elib=19436 where...

 

[solderdude]

When they really did EQ the bass to Harman 2013 target then the bass in the OE headphones would already have a +4dB boost and any adjustments the testers made would have been relative to that.
The average level people prefer to listen to seems to hover around 70dB so lower than when the sound was mixed/mastered so requiring some 4dB lower bass boost to get the same 'tonal balance' as it existed in the studio so it is not strange to me that such a low bass boost is needed to sound 'good'.

 

[HRTF_Enthusiast]

When they really did EQ the bass to Harman 2013 target then the bass in the OE headphones would already have a +4dB boost and any adjustments the testers made would have been relative to that.

The EQ didn't include a bass shelf.

This is confirmed here:

 

[solderdude]

2013 HP curve would be a +6dB in the low bass (blue trace bottom plot) and the result would thus measure like the upper plot in your post.
+6dB is a bit too much for me.
I prefer +4dB and acc. to equal loudness contours difference between 80 and 70dB SPL instead of the sharp tone control Harman used.
This gives me roughly the equivalent of tonality in a studio using (relatively) flat nearfield monitors at 80dB average SPL.

So Harman-esque in essence but less boost and following a more natural curve.

 

[HRTF_Enthusiast]

2013 HP curve would be a +6dB in the low bass (blue trace bottom plot) and the result would thus measure like the upper plot in your post.

I'm not sure I understand you. If 2013 HP curve has a bass boost and the top plot has an invisible bass boost, why did the author say the mean bass gain of 2.89 dB is on the low side?

 

[MayaTlab]

I have used DT990. Admittedly, I haven't used it with EQ, but the distortion was not enjoyable.

Since I believe that you have access to AES : https://www.aes.org/e-lib/browse.cfm?elib=17441
The audibility of non-linear distortion is not superbly determined yet, but in the above study two headphones were equally preferred, when excluding FR as a variable, despite having two very different distortion profiles.

As a general rule I suspect that what most people think is "distortion" is just FR, in ways that aren't always expected or difficult to grasp (for example if the on-head response you're experiencing is quite different from the results obtained from ear simulators).

 

[solderdude]

I'm not sure I understand you. If 2013 HP curve has a bass boost and the top plot has an invisible bass boost, why did the author say the mean bass gain of 2.89 dB is on the low side?

The top plot is what was measured including the Harman curve so bass would be lowered in that plot and treble is different looking from the Harman curve as well as it deviates there too but not as obvious, at least that is my take on this.

My assumption though as I never read the actual paper thoroughly as I did my own, extremely limited and only personal research.

 

[HRTF_Enthusiast]

The top plot is what was measured including the Harman curve so bass would be lowered in that plot and treble is different looking from the Harman curve as well as it deviates there too but not as obvious, at least that is my take on this.

But if the bass was accurate to the 2013 Harman Target, that would contradict this statement:

 

[solderdude]

No idea what they mean by that.

The only thing I can figure is that the 2013 Harman bass boost is about +6dB.
I found that a low bass boost of +4dB sounds best to me (at sensible levels one can listen to for hours).
When they found mean bass level boost was almost 3dB and they find that a bit on the low side then either the average bass boost was 3dB below Harman (so +3dB, close to my +4dB) or it was +9dB which seems unrealistically high.

 

[HRTF_Enthusiast]

or it was +9dB which seems unrealistically high.

They said 2.89 dB was on the low side of 2013 Harman target, so not possible.

When they found mean bass level boost was almost 3dB and they find that a bit on the low side then either the average bass boost was 3dB below Harman (so +3dB, close to my +4dB)

This is what looks to be the case

 

[Robbo99999]

I'm not sure I understand you. If 2013 HP curve has a bass boost and the top plot has an invisible bass boost, why did the author say the mean bass gain of 2.89 dB is on the low side?

They shouldn't have called it the 2013 Harman Curve - it's not. God knows why they did, but it's confusing the conversation here.

They shouldn't have called it the "2013 Harman Target" in this post (in Fig 4):

Headphone Measurements Using Brüel & Kjær 5128 HATS

That paper is quite misleading. This paper? Although I do have some reservations about it, I wouldn't say it's misleading, it just uses a different method (cluster analysis based on headphone frequency response preference) of determining bass preference compared to the tone control...

www.audiosciencereview.com

 

[GaryH]

If they used the flat in-room response, isn't the mean bass gain level basically flat on 5128?
View attachment 166536

At less than half the bass level of the Harman target, the results from that study are inconsistent with previous MOA preferred bass level studies by Sean Olive with many more listeners, which puts them in doubt. For example:

Maybe it was methodological issues, or maybe the distortion of the DT990 did play a role here. In this study by Sean Olive and Steve Temme, one headphone was rated significantly lower than the others (all EQed to the same target) due to its distortion, notably high below 100 Hz:

(THD measured at 82 dB, 88 dB, 94 dB and 100 dB)

Granted, this is higher than the DT990, but I don't think it's outside the realms of possibility that the DT990's relatively high bass distortion could have influenced listeners' bass level preferences, as pushing it too high may have resulted in perceived diminished sound quality due to an increase in this distortion to audible levels (maybe only during transient peaks). Incidentally, this is possibly why Oratory doesn't push the sub-bass of the DT990 all the way up to the Harman target in his EQ profile for the headphone.

The data could be made a lot more useful if we could see which listener IDs were trained .