Convolution Filter for AKG N40

I’ve prepared a convolution filter for the AKG N40 for anyone who’s interested. I took the measurements from Head-Fi (graph) and run this through WebPlotDigitizer and AutoEQ.

Compensation:


Filter:

Download AKG N40 filter.

Edit: Updated to use correct compensation.

I realize the following may sound a bit negative. Be assured that I do not mean it that way; it’s just my clumsiness with the English language :slight_smile: !

I believe I’m either missing some info, or misunderstanding it. I have also seen your Beyerdynamic topic.
To the naked eye, this seems to be a sequence of a small handful of eq sections. Two or three sections at most, I’d say. Perhaps even 1 could do.

What’s the reason for using convolution here ? It feels a bit like firing up your supercomputer, to calculate the square root of 9 … :slight_smile: .

Second. It seems clear, that you are optimizing against ‘some’ target response. What exactly is that target, and why do you (or anyone) consider that to be optimal ?
I do not quite understand; especially since the target for the Beyerdynamic seems to be a very different one ?

Nice to see you play around with it anyway ! You reminded me : 'ah. yes. need to play around with the crossfeed, when I find the time… :slight_smile: '.

You cant load in parametric EQ from other software so your limited to entering manually which is a but dull. Exporitng the filter as convolution allows you to simply import the filter.

Okay, thanks. That’s one answered, and one still to go :slight_smile: .
What was the target for optimizing against ?
In case you wonder why I’m asking : I would normally expect such a correction, to correct something that is universally ‘wrong’ with these headphones.
And not an adjustment to taste, which seems to be the case here somewhat (especially since both targets appear to be different). And which are relatively easy to implement with additional EQ, as apposed to anything that is really ‘universally wrong’.

It may very well be that I’m misunderstanding something; really not saying that anything is wrong.

AutoEQ is a project for equalising headphone frequency responses. In simple terms, the objective is to correct the headphones to a neutral sound. This is why the Beyerdynamic and AKG have different graphs.

The convolution-based filters provide settings for equalisation applications including Roon and comprise parametric equaliser parameters, standard 10 band graphical equaliser levels and impulse responses for convolution based equalisers. Roon uses the latter.

I created the filters because they weren’t available in the AutoEQ library. I did this by locating the frequency response graphs for the headphones—I acknowledge that I am trusting the person who prepared these for doing a good job—and feeding this into WebPlotDigitizer to extract the numerical values.

I think I got this bit right. :slight_smile:

Finally, I run the data though AutoEQ to create the filter: this is provided as two WAV files. One for 44.1kHz and the other 48kHz.

There’s more at AutoEQ: https://github.com/jaakkopasanen/AutoEq/blob/master/README.md.

I did this as an experiment. The Beyerdynamic is a good headphone and it didn’t perceive a difference after listening for about an hour. However, with the AKG I could tell the difference straight away. In my opinion an improvement.

Thanks, appreciate your answer!

That’s a surprising result to me. These two corrections really do not differ that much. Yes, the correction for the Beyerdynamic is slighly more wideband. And thus expected to be a bit more audible. Probably more ‘punchy’. But the difference is quite small… I would not expect one to be obvious, and the other not. Nice obversation nonetheless :slight_smile: . EDIT : I now realize that you meant the opposite. Sorry. That’s even more surprising to me :stuck_out_tongue: .

I have highlighted the parts where I would like additional clarification.
I do have in-depth knowledge on the Harman research on the way headphone sound is perceived, but I do not know too much about other researchers.
Is it possible for you to link to (or explain how) a source that explains how neutral response is achieved in headphones, according to the principles used in AutoEQ ?
(I have searched, but not found much useful by myself thus far…).

Again : pure interest, no criticism :slight_smile: !

I’ve just checked and I think I made a cut ‘n’ paste error in the compensation settings, so will return to this over the weekend when [I’m] back home. So maybe I will have a different view then. For now I will remove the filter links.

It’s my understanding that the neutral sound is achieved by compensating for the Harman target response with adjustments for over-ear, on-ear and in-ear headphones. This is achieved by calculating the difference between raw microphone data (the curve I put through WebPlotDigitizer) and the compensation curve, i.e. Harman. Probably an oversimplification, so you may want to read https://github.com/jaakkopasanen/AutoEq#technical-challenges.

I think the other researchers either provided the data processing or measurements for the numerous headphones listed in the Results folder on GitHub.

Maybe you could share your knowledge about the Harman target? I understand that this is used in mixing to provide sound engineers with the sound like good speakers in the home.

Thanks again for responding. I can see that you’re being careful - I cannot express how highly I appreciate that!

In my timezone, it is now time for bed. I will make sure to investigate your link and explanation in the next week, and report back to you. Thanks again so far.

1 Like

Well, that was a lot simpler than expected - it was my own error :). My bad :slight_smile: .
I assumed this AKG to be an over-ear set. But it´s an in-ear. Having a different target for in-ear, and over ear makes perfect sense to me.

Still, I like to point out that it seems odd that your files only make some corrections below 100Hz-ish.
Either both headphones do really well in tracking the ´optimum´ targets, or something is amiss in the correction ,-).
I haven’t thoroughly read up on AutoEQ yet, so I don’t know if it could cause issues like this. I doubt it.

Next up, your question on Harman. Prepare for a lot of text. Brevity is not one of my virtues…

Your understanding on the Harman target is indeed mostly correct. A forum post will not be enough to cover it, but I´ll try to give a little glimpe.
Ideally you would be an AES member (to read their papers published there), and you have prior knowledge on their research on preferred in-room responses for loudspeakers.
No, I haven’t read every single paper either ;-).

The target is not ´set in stone´. As research continues, the target is adjusted according to new findings.
It also allows for some variation, due to varying listener`s preferences. Especially in the lows. (I guess you now understand why I´m a bit confused by your correction files?).

The target aims to be a response, that is generally preferred by trained listeners. Additional research correlates that level of preference, to actual neutral delivery/reproduction.

A bit too simple : The target aims the perceived headphone sound, to mimic that of a set of in-room loudspeakers with the pre-determined ´ideal´ response (hence your need for prior knowledge on that).
It is also explained why no comparison is made with the ´original live sound´ - covering this as well is a bit too much for this topic.

I believe this last paragraph is what you meant, when saying :

So yes, it is designed to mimic your perception of an ´ideal´ loudspeaker. But no, it is not designed with (only) sound engineers in mind.

I think it´s good to stress that this target is not the ultimate answer. First, it is likely not finished.
And second, there are other schools of thought, with slightly different, but equally valid findings.
But yes, it is the most thorough research that we have by today. Amazing work.

If really interested to dig a little deeper, these could be a nice starting point :

Now, excuse me while I´m replacing a few worn keys on my keyboard :slight_smile: .

I checked my bash history and indeed I had made a mistake and referenced the wrong file for compensation. I’ve corrected this and updated the original post for both AKG N40 and Beyerdynamic T5p 2nd Gen. I’ve also included the Harman data.

Thank you for the explanation and links. It’s also worthwhile noting that Harman update the compensation files periodically. I’ve used the most recent and used the files without bass boost.

Ha, that looks a whole lot more logical now !
However, this is almost the opposite of what you had previously - a very severe bass reduction/mid lift, instead of slight bass lift. Have you listened yet ?
I’d expect this correction to sound much thinner, and perhaps not an improvement considering your previous judgment. But I cannot know for sure from behind the keyboard ;-).

I have to say : I have trouble understanding unambiguously what the colors in your graphs mean. I should dive deeper into AutoEQ, first.
I should also mention : It is my understanding that Harman applies a correction on their own measurements, and that this correction is not (yet) universally accepted as a standard. If indeed so, this could very well lead to a mis-correction (if ‘your’ measurement indeed does not meet the Harman standards)

I’ll await your experience first, before commenting on that.

Ah, that’s because the input file and compensation were the same. Doh! Although I haven’t listened to the N40s yet I’ve tried the T5p 2nd Gen.

Anyway, you are correct, the sound is thinner although volume level differences somewhat exaggerate this effect. Overall I prefer the clarity, but I don’t think it’s quite right. I may revisit the graphs over the weekend and see if I get the same results second time around. It could be that a reliance on someone else’s measurements is the issue and as you say there could be a calibration issue.

Also, I noticed that the Audio-Technica ATH-M50 are precompiled on the AutoEQ website. I’ve tried this filter and while the sound is thinner than without it is a marked improvement.

I’ll give the N40s a go tomorrow and also look at building a parametric equaliser with the data.

Signing off now.

I know it’s kinda an old topic but, do you have the file for AutoEQ? Just realised that the database doesn’t include the AKG N40

1 Like