Beyerdynamic DT 880 600 Ohm Review (Headphone)

[solderdude]

I agree with what oratory wrote.
My point is this: when all DT770-250 differ from the average and you create an 'exact' EQ (multiple bands) for that average then probably none if the DT770-250 will sound 'accurate' as all of them differ audibly. They will all sound equally different but just with a better tonal balance then before.

Sonarworks also measures lots of headphones and investigated the HD650 with models over the years and found that within a certain production year they measured closely the same but different over the years (for a certain period). Those that will measure HD650 in the coming years will also measure differently because of the pad change (same driver). I hope suggesting an 'average' EQ is not the answer when differences are substantial.
In that case all HD650 will sound 'wrong' when corrected.

 

[Maiky76]

Well even for Oratory's original DT880 measurements with a pre-EQ score of 67, his post-EQ score was 99, a significant 17 points higher than your algorithm's



He'd probably show that individual data for you if you asked him. However, unless you have measured many units of each production year and found a significant common pattern, you've likely just seen unit to unit variation. As Oratory says:

The two points that you are missing:

The points I was trying to make might have been lost, sorry:
- The data used for the EQ design is (obviously) critical and initial scores 68/88 discrepancy is very significant, so comparing the EQed scores as you did is IMO irrelevant .

No matter fact the scores are similar the trough at 10K on the ASR data is much deeper. Without EQing it it will ALWAYS drag the score down both for the standard magnitude deviation AND the slope when it becomes the driving factor in the score, i.e. when the rest is properly EQed.
It similar to the proverbial glass ceiling....

- One can practically always find a way to EQ differently and one-up the EQ that is tested regardless of the suitability/sanity of the "score based improvements". So yes, if I did not compensate for the 10k notch, the score would have been about 106 (after re-optimization).
I did not do so in the original EQ because although I knew it would help the score, it might not have yielded the best performance...

Well even for Oratory's original DT880 measurements with a pre-EQ score of 67, his post-EQ score was 99, a significant 17 points higher than your algorithm's

I just ran my algorithm on their (scanned) data and their EQ. I got a better score.

Score no EQ: 66.9
Score Original Oratory EQ: 98.3 -> their ceiling is higher thanks to the smoother FR
Score Tweaked Oratory EQ: 100.3

Does it make this EQ better?
Probably not but it does show that only looking at the score is not correct.
Why not make a FIR and have the perfect score i.e. 114.490443008238?

What does it prove?
My algorithm is at least as good as their's otherwise the score would have decreased.
As a matter of fact it is easy to cheat the metric to get a higher score still, without EQing the 10k notch with the same EQ structure.
However I doubt that the actual listing test would agree.

I think you should try to EQ based on the score yourself to understand how the metric is working...

End of story I think.

 

[Robbo99999]

The two points that you are missing:

The points I was trying to make might have been lost, sorry:
- The data used for the EQ design is (obviously) critical and initial scores 68/88 discrepancy is very significant, so comparing the EQed scores as you did is IMO irrelevant .

No matter fact the scores are similar the trough at 10K on the ASR data is much deeper. Without EQing it it will ALWAYS drag the score down both for the standard magnitude deviation AND the slope when it becomes the driving factor in the score, i.e. when the rest is properly EQed.
It similar to the proverbial glass ceiling....

- One can practically always find a way to EQ differently and one-up the EQ that is tested regardless of the suitability/sanity of the "score based improvements". So yes, if I did not compensate for the 10k notch, the score would have been about 106 (after re-optimization).
I did not do so in the original EQ because although I knew it would help the score, it might not have yielded the best performance...

Well even for Oratory's original DT880 measurements with a pre-EQ score of 67, his post-EQ score was 99, a significant 17 points higher than your algorithm's

I just ran my algorithm on their (scanned) data and their EQ. I got a better score.

Score no EQ: 66.9
Score Original Oratory EQ: 98.3 -> their ceiling is higher thanks to the smoother FR
Score Tweaked Oratory EQ: 100.3

View attachment 139767

Does it make this EQ better?
Probably not but it does show that only looking at the score is not correct.
Why not make a FIR and have the perfect score i.e. 114.490443008238?

What does it prove?
My algorithm is at least as good as their's otherwise the score would have decreased.
As a matter of fact it is easy to cheat the metric to get a higher score still, without EQing the 10k notch with the same EQ structure.
However I doubt that the actual listing test would agree.

I think you should try to EQ based on the score yourself to understand how the metric is working...

End of story I think.

Yes, there's no point obsessing about the score, I think your EQ's are good from looking at the frequency response charts, just don't EQ up that 10kHz natural dip which you did mistakenly EQ up one time. I am certainly a fan of computer/graphical based EQ'ing whether that be your algorithm or me using REW to manually EQ to a measurement or Oratory's EQ's that are also graphically based.....I think that's far more reliable than just eyeballing the filters & frequency response like Amir does, you can see in his K7XX EQ that he was well off the mark (and for no good reason) on the bass EQ, and bit on the treble too.

 

[markanini]

I'd go a far as saying don't EQ above 2Khz due to unit and fit variation being the bigger factor. But leaving 10kHz alone is a reasonable minimum.

 

[GaryH]

End of story I think.

Not quite!

What does it prove?
My algorithm is at least as good as their's otherwise the score would have decreased.

Disagree, because as you yourself say...

Does it make this EQ better?
Probably not but it does show that only looking at the score is not correct.

There are other advantages to Oratory's EQ profiles:

1. Because he initially matches the measured frequency response to the target over a range instead of a particular frequency, the max filter amplitude and so negative preamp magnitude required for his EQs are usually lower e.g. around 3 dB less for the DT880 compared to yours, which results in more volume headroom, and so allows the EQ to be used on a wider range of playback devices for which the user may be approaching max volume post-EQ.

2. He listens to the headphones post-EQ and fine-tunes the EQ based on issues that may not be immediately visually obvious from the frequency response.

3. His newer EQ profiles include carefully chosen filter frequencies and bandwidths to correspond to intuitively understandable regions of the frequency response to not only adjust for taste, but to account for unit variation. From his latest DT880 pdf:

Adjust gain of band 1 to preference (bass)
Adjust gain of band 2 to preference (warmth)
Adjust gain of band 3 to preference (tonal accuracy)
Adjust gain of band 7 to preference (treble detail)

Auto-generated EQ profiles such as yours and AutoEQ's are great and much appreciated, but in my opinion a carefully made manual EQ such as Oratory's will invariably be better, for the above reasons unrelated to the post-EQ score. (Oh and I suggest you only report scores to the nearest whole number, as the decimals really have no audible significance.)

 

[GaryH]

I agree with what oratory wrote.
My point is this: when all DT770-250 differ from the average and you create an 'exact' EQ (multiple bands) for that average then probably none if the DT770-250 will sound 'accurate' as all of them differ audibly. They will all sound equally different but just with a better tonal balance then before.

And that bolded part is precisely the point of EQing: using Oratory's EQs will bring a particular unit on average closer to the target, improving its tonal balance. Point 3 above can then come into play if fine-tune adjustments need to be made due to unit variation. Unless the user has an industry standard HATs, this is the best that can be achieved with headphones with poor unit consistency such as these Beyerdynamics. Using an EQ based on one particular unit will for the majority likely produce worse outcomes than an EQ based on an average, unless for example the production year of that particular measured unit is known, the production year of the EQ user's headphone is known, and there is data showing definitive correlations between production year and frequency response, which I have not seen for the Beyers. Or do you have such data you can share with us? (Data on the HD650 is irrelevant as unit variation among manufacturers and models differs wildly.)

 

[Maiky76]

There are other advantages to Oratory's EQ profiles:

1. Because he initially matches the measured frequency response to the target over a range instead of a particular frequency, the max filter amplitude and so negative preamp magnitude required for his EQs are usually lower e.g. around 3 dB less for the DT880 compared to yours, which results in more volume headroom, and so allows the EQ to be used on a wider range of playback devices for which the user may be approaching max volume post-EQ.

Two lines of warning I put in my all my headsets EQs:

• I am using PEQ (PK) as from my experience the definition is more consistent across different DSP/platform implementations than shelves.
• With some HP/amp combo the boosts and preamp gain need to be carefully considered to avoid issues

Regarding normalization I do the same, but I do NOT only eyeball a Least Mean Squares fitting which is what I believe you are referring to.
I also minimize number of biquads needed to fit the curve which may result in a not so trivial normalization frequency point.
People do not pay attention of the biquad definition but from experience they vary widely so even you input the correct parameters it does not guaranty that the actual EQ is as designed in particular with Shelves filters. I also cater for specific requests when nicely asked.
This enables a wider range of device to be used.

Your volume point is moot: in my original EQ the Preamp gain is set at -7.7dB and Oratory's -9.8dB.
This 100% determined by the max boost of the EQ nothing else for any consistent EQ. It is a zero sum game.
If you expect to be able to boost any frequency by 10dB as an example and have no impact on the dynamic range of the system (i.e. Max SPL) you are wrong.
EQing is trading dynamic range for accuracy. No free lunch I am afraid, unless the system has a large headroom built in as in: it will handle/produce much more power than required to achieve desired max SPL.

SNR can be different but it is a per case per case basis issue (i.e. what produce the most noise in the gain chain) no general rule applies.

To close the loop what do make of this, using the ASR data, my original EQ vs Oratory's original EQ (both base don similar score (68/67).
and the supposedly superior Oratory's EQ:

All your objections should work both ways, should not they?

3. His newer EQ profiles include carefully chosen filter frequencies and bandwidths to correspond to intuitively understandable regions of the frequency response to not only adjust for taste, but to account for unit variation. From his latest DT880 pdf:

Oratory's:
Adjust gain of band 1 to preference (bass)
Adjust gain of band 2 to preference (warmth)
Adjust gain of band 3 to preference (tonal accuracy)
Adjust gain of band 7 to preference (treble detail)

What is stated here is quite tautologic in essence: if you change the system will sound different.
In particular band 3, I though we were looking for accuracy so I don't see how it can have different settings.
This band it is either accurate or not....

One more line of warning I put in my all my headsets EQs:

• The EQs are starting point and may require tuning (certainly at LF).

I cannot predict the users taste so up to them to tweak. Expect a learning curve though.
I sometimes use Amrim's preference (documented as more bass and less HF) and modify the Harman curve to show that for him his EQ make sense. but that's all.

As a general comment my algorithm does not simply optimize for the curve fitting nor the score.
There is way more going on in the fitness function of the evolutionary algorithm and it is a WIP and it has been for years.
I welcome any feedback, good or bad, in order to improve it.

Oratory know what they are doing, there is no denying that, but it does not mean that others can't do a similar job.
We'd better leave it at that.

 

[GaryH]

Your volume point is moot: in my original EQ the Preamp gain is set at -7.7dB and Oratory's -9.8dB.

That's not a like-for-like comparison. When comparing Oratory's and your EQ profiles both for his data, his preamp is -5.4 dB and yours is -8.2 dB. I was thinking about problems arising when using devices that lack a dedicated preamp setting e.g. the miniDSP IL-DSP, which only has a physical digital main volume control. Your EQ would reach digital clipping at a lower volume on such a device, therefore Oratory's EQ would provide more clean volume headroom.

To close the loop what do make of this, using the ASR data, my original EQ vs Oratory's original EQ (both base don similar score (68/67).
and the supposedly superior Oratory's EQ:

View attachment 139969

Again, a misleading comparison - similar scores are not necessarily indicative of similar frequency response. Of course an EQ profile developed for one specific unit will likely be better for that unit than one developed for an average of several units (especially as this average doesn't include the unit in question). But that's not very useful for the reader, as their unit may differ from this specific one. What is more useful for the average user is, naturally, an EQ based on averaged data, and the more units included in that average the better (i.e. his latest pdf, not the earlier one with fewer units you're referring to here).

What is stated here is quite tautologic in essence: if you change the system will sound different.
In particular band 3, I though we were looking for accuracy so I don't see how it can have different settings.
This band it is either accurate or not....

I thought this was clear. The degree of tonal accuracy heard is dependent on three main variables:

1. How close to the average a particular headphone is (most important for this headphone due to its high unit variation).

2. The mixing/mastering of the recording being listened to (circle of confusion issue).

3. The anatomy/hearing of the user.

Fine-tuning this band (and the others Oratory highlighted) allows the listener to adjust for these variables.

 

[solderdude]

4. Seal, hair between driver/ear

5. pad wear

6. clamping force/head width

7. positioning on the head.

8. Listening volume

9. taste

 

[Sharur]

4. Seal, hair between driver/ear

5. pad wear

6. clamping force/head width

7. positioning on the head.

8. Listening volume

9. taste

Would the pads molding around the glasses reduce bass roll-off?

 

[Sharur]

bass feels somewhat stronger without glasses but this is obviously impossible to blind test

 

[Sharur]

HD800S. Not because of tonality but because of the very nice spatial qualities of the 800S where it almost separates instruments into layers.

because of immense midrange group delay?

 

[Maiky76]

That's not a like-for-like comparison. When comparing Oratory's and your EQ profiles both for his data, his preamp is -5.4 dB and yours is -8.2 dB. I was thinking about problems arising when using devices that lack a dedicated preamp setting e.g. the miniDSP IL-DSP, which only has a physical digital main volume control. Your EQ would reach digital clipping at a lower volume on such a device, therefore Oratory's EQ would provide more clean volume headroom.

I beg to differ, what you hear is SPL. We can compare since the target curves are identical and assuming we are EQing the same sample.
It all depends on the normalization frequency.
I normalize in the low midrange Oratory at HF which set the sensitivity of the system lower if we normalize at "my" frequency.
Oratory's EQ magnitude is on average 2.6dB lower than mine at same voltage input.

Preamp gain for Oratory is -5.4, -5.4-2.6 = -8dB, it really should be 8.2dB+ as it is when similarly normalized to my frequency the max EQ is 8.2dB+.
My EQ will clip at -8.2dB so the same SPL
You just cant look at the preamp gain and say that the only important parameters, SPLs at max output, will be different in this case they won't.
Just repeating myself:

This 100% determined by the max boost of the EQ nothing else for any consistent EQ. It is a zero sum game.
If you expect to be able to boost any frequency by 10dB as an example and have no impact on the dynamic range of the system (i.e. Max SPL) you are wrong.
EQing is trading dynamic range for accuracy. No free lunch I am afraid, unless the system has a large headroom built in as in: it will handle/produce much more power than required to achieve desired max SPL.

Again, a misleading comparison - similar scores are not necessarily indicative of similar frequency response. Of course an EQ profile developed for one specific unit will likely be better for that unit than one developed for an average of several units (especially as this average doesn't include the unit in question). But that's not very useful for the reader, as their unit may differ from this specific one. What is more useful for the average user is, naturally, an EQ based on averaged data, and the more units included in that average the better (i.e. his latest pdf, not the earlier one with fewer units you're referring to here).

Misleading? Again? Don't you think your are exaggerating a bit there?

We start with similar scores 68.6/66.9 (single sample no averaging), different sample but identical product, similar measurement apparatus etc.
I can't see why this comparison is not valid. End users will not get any of the measured units so we are in the exact boat as anyone buying a DT880.
You brought the magic 106 score to the table naively yourself, when one digs a bit, it just does not make much sense to compare as demonstrated.
You said it yourself one should not compare two different units data/EQ as you did more so if one is averaged.

To illustrate:
- Oratory average based EQ on ASR sample: 66.8 from 68.6
- Oratory average based EQ on Oratory first sample : 80.1 from 66.9
- Oratory average based EQ on Oratory averaged data: 106 from 88

Honestly, the average user should be totally lost there and the EQ does not look useful under this light: the same EQ yield totally different results on two different samples and an averaged data especially since most people think the average data should yield better results across the whole production. For this particular product tolerance seems awful.
Designing EQ based on average or averaged EQ is actually really difficult. One should take into account the standard deviation of the population. amongst other things....

Fine-tuning this band (and the others Oratory highlighted) allows the listener to adjust for these variables.

My comment still stands, I see no compelling argument in what you wrote.
Olive study showed that the default Harman curve bass (-3/+6dB) and HF (+/-1.5dB) works for the vast majority of people including their taste, their age/gender group training level etc. There is no "tonal accuracy band" that was revealed, even less so one that will set accuracy by itself and solve your three, correctly stated, points in particular the circle of confusion that we cannot escape...

In addition I noticed that Oratory states that the scores are normalized to 100, still the averaged EQ yields 106.
a. The normalized preference ratings are used, where zero deviation from target equals a preference rating of 100
b. 106
The perfect score, where zero deviation from target, is around 114. so one of these two statements is necessarily wrong.

 

[solderdude]

because of immense midrange group delay?

Because of the driver ear distance, inner cup dimensions, and driver angle as well as the driver membrane shape.
I have not ever seen any relevance between group delay and imaging. It is only good for spotting pad bounce and narrow resonances IMHO.

Would the pads molding around the glasses reduce bass roll-off?

HD800 is hardly affected by seal issues. Of course when the arms do not rest against the skin where it leaves the pads there will be some leakage there.

 

[Sharur]

Because of the driver ear distance, inner cup dimensions, and driver angle as well as the driver membrane shape.
I have not ever seen any relevance between group delay and imaging. It is only good for spotting pad bounce and narrow resonances IMHO.

 

[solderdude]

No it doesn't. Someone coined it that it MAY have something to do with it and Amir looked into this.
Not all open headphones that showed this did have good spatial qualities and closed headphones that do show good spatial qualities do not show this 'noise'. There is no relation between the noise and spatial qualities.
This type of noise is seen in nearly all measurements of open headphones. IMHO it is likely caused by background noises being picked up.

 

[Robbo99999]

Regarding normalization I do the same, but I do NOT only eyeball a Least Mean Squares fitting which is what I believe you are referring to.
I also minimize number of biquads needed to fit the curve which may result in a not so trivial normalization frequency point.

That's a good point, I noticed the same thing regarding the manual EQ's in REW that I do for headphones.....choosing the right overlap point (where the target sits on the measurement) can make the EQ easier resulting in less number of filters. (I eyeball where to line up the target on the measurement, but choosing the right place can still make it easier to EQ to the target, and with less filters.) The overlap point for me often ends up being in & around 1kHz, but not always the case.
EDIT: for example, here's one that I decided to align at 5kHz, so this was one of the more unusual places (an HD600 EQ from an Oratory measurement):

 

[Sharur]

No it doesn't. Someone coined it that it MAY have something to do with it and Amir looked into this.
Not all open headphones that showed this did have good spatial qualities and closed headphones that do show good spatial qualities do not show this 'noise'. There is no relation between the noise and spatial qualities.
This type of noise is seen in nearly all measurements of open headphones. IMHO it is likely caused by background noises being picked up.

that statement was a quote from amir, not me

 

[solderdude]

Yes, I explained the how and why.

 

[Sharur]

Yes, I explained the how and why.

so how are spatial qualities explained?