Headphone Measurement Stuff

Here are Audeze's results with bypassed ear canal (mic inserted into open ear canal, ear canal entrance not blocked): LCD4 and HEK

and with a mic close to the ear drum: LCD4 and HD800


Obviously you now get the 10db ear canal gain at 2.5kHz. Of course the outer ear does have some effect on the FR (the graphs I posted in my thread), but I feel that it's actually a more realistic method for the upper mids, mainly because the difference (between bypassing the outer ear and not) is never the same for different headphones.

 

Was about to ask, thx. This is closely related to what I wanted to get at in the other thread, but I don't think it's possible to merge them.

That's what I did! I compared my IEM measurements and the coupler measurements for various headphones to see exactly how much gain the outer ear adds and at what frequency in this post. I also did the same thing for speakers. The plots only look weird because I intentionally did it the other way around, because I think the flat at the ear canal opening plots are a good target for the coupler measurements. To get the outer ear gain you have to mirror them vertically.

Also the mic I have isn't small enough to go all the way into my ear up to the eardrum. That's why I quoted Audeze's measurements, for the comparison between eardrum level and ear canal opening with a real ear.

 

Hahahahhaha

@ultrabike is going down the same path of confusion that I did when @Serious tried to explain his convoluted setup to me, almost step for step.

 

@ultrabike the plots are created by compensating the coupler FR with the in-ear-mic FR, giving you the difference between both. If you were to do it the other way around, you would obviously get the same plot but inverted. I only did it inverted because @Marvey said the B&K target worked well for his coupler and I wanted to show him what I thought was a better target response and the problem with the coupler that the target response is different for every headphone. The differences frequently exceed 3db, which is why I think the coupler is inaccurate. The ear simply doesn't change the response in the same way for every headphone.

The speaker plots are a little more difficult because you can either play both or only one speaker which will change the response noticeably. After I arrive at one plot for the omni mic speaker response and another plot for the in-ear-mic response (after some special sauce treatment), I again compensate the omni mic response with the in-ear-mic results and then get a plot that is similar to the one posted in the thread. But because flat isn't a good speaker target IMHO, I apply an inverted B&K curve compensation to the difference plots to get rid of the rising response ( -(-B&K) = +B&K).
(I do all of this compensation stuffs to get rid of the speaker FR, even if they're usually pretty neutral.)

What this also means is that the B&K target will make speakers have a very flat FR from 20Hz to 2kHz and from 6kHz on at the ear canal opening, which to me means the B&K target is a very good speaker target. But I don't think the B&K target is simply the inverse of the effects of a head-shaped sphere on the FR. As Marv and a few others said it's probably a good idea to have a small dip around the outer ear gain region, depending on the angle of the speakers to the head, speaker directivity, diffuse room reverb, and many other factors. I do like a small dip around 3kHz under most circumstances, but I don't compensate my speaker measurements for this. I usually do however compensate the omni mic measurements with the B&K target.

 

Music is a mix of mono and stereo information. The waves add differently with a real head. For omni mics, yeah, use one speaker only, but for in-ear-mic measurements I prefer to use a mix of both, using only one speaker and using both speakers at the same time. This is also a decision Tyll struggled with. I honestly don't think both are accurate.

To be blunt, I only use my weird in-ear-mic setup because it's more accurate to what I hear. I almost gave up on measuring the HD800 until I found out that the 4kHz dip was mainly an artifact of flat plate couplers that doesn't appear on a real head (or a dummy head for that matter).
I simply don't hear the HD800, Elear and HD600 like this, I hear them more like this. No compensation curve is going to get the coupler measurements where I feel they are accurate.

 

@Marvey I also prefer the V1 measurements for CSDs above all other measurements. That's why I searched for a material that made my stock HD800 measurements look nearly identical to yours, but without the need for a flat plate. (Actually searched is saying too much. I stumbled upon it. I don't have all that much left. Maybe enough for one or two more couplers.)
Also of course I don't think the in-ear-mic results are the be-all-end-all. There will never be such a thing with measurements (no matter what kind). I use the V1 for treble FR (above about 5-6kHz) and CSDs. I actually find the V1 type measurements more "accurate" than V2. In my case I could "compensate" the results to go from V1 to V2 FR with the HD600 and HD800, so I feel that all flat plate couplers generally interact with headphones in a similar way, just with different damping and loss properties (to oversimplify it a lot). Of course compensation won't get you very far when the headphones have very different radiation properties, pad design, etc. I'm trying to find a coupling material that is closest to the skin on my head in this sense. While this will still not get you ultimate FR accuracy, I have a feeling that it will probably get the CSD results to be more realistic. It's probably somewhere in between V1 and V2.

 

Again this depends on what you're measuring with. The FR effects are different for the head than for an omni-mic. And music is both stereo and mainly mono.

My in-ear-mic response is not compensated. The only compensation is for the mic response like everyone does, but the mic is very flat. The position sensitivity isn't so big of a deal since you know where you're supposed to put the headphone on your head unlike with a coupler. But yes, results are very much more position sensitive than with a coupler and they should be. I'm still not a fan of averaging 5 or more plots like Tyll does since I always put my HD800 at the same spot on my head. It's pretty precise, down to less than 3mm or so.
The HD800 plots are the HD800X. I've posted my stock HD800 in-ear-mic plots somewhere else. They did indeed look crazy bright and peaky with recessed upper midrange. Even the other HD800s in the thread and the HD800S did.

Yes, I said the results above 5kHz or so are inaccurate and it's probably better looking at the coupler plots for that region. Like I've said before, I don't hear any of these cans as severly rolled off above 10kHz. For whatever weird reason the HD800 measures the most rolled off but actually sounds the airiest to me, so yeah, accuracy goes to shit above 10kHz.
I was mainly talking about the upper midrange response. The whole response from 500Hz to 4kHz on the coupler doesn't match my subjective impressions.

Also the hair affecting a perfect seal argument is moot IMO because most people have some hair affecting the perfect seal. My hair isn't all that long and I've not seen huge changes when I cut my hair, or even with and without my glasses. I actually tried to measure the effect of hair and it wasn't as great as the felt for example. If anything it increases the accuracy I think.


I forgot to mention this, but one thing that is very important is the microphone insertion depth. You want it to be just flush with the ear, otherwise you get too much treble in my experience. This actually happened with the HD800X plots. Like I said in the thread, the region above 6kHz is usually on the same level as 500Hz.

When I talked about compensation it was about the difference plots, which are not FR plots of headphones. Those are the plots I linked to in the other posts.

 

I said I hear them "more like this", not just like that. Of course you get the ear resonances that you don't necessarily hear like that. That's why I like using the flat plate coupler for the treble and also why I don't post the CSD plots for the in-ear-mic measurements. The 5kHz bump might also be overdone on both the HD600 and HD800 and I don't think I heard the sharp 6kHz spike on the Elear.

 

Some caveats:

1) I do not think "air" necessarily correlates to >10KHz. In fact, from my experience, any emphasis above this frequency range is extremely annoying depending on your age. I personally can still hear up to 19KHz, so any emphasis in this range sounds super piercing and annoying to me. Sometimes they even make the upper mids "shouty" or "harsh" because of harmonic contents... (3rd harmonic of 4KHz is 12KHz, btw...) even when the upper mid levels do not look crazy. Closed-back headphones generally do this to me... they sound super piercing/harsh even when they aren't bright. So it is best to separate the distinction between >10KHz region and "air". Or at least the correlation there doesn't apply to me.

2) I have used in-ear mic, flat-plate couplers, etc... a lot, and as said, consistency is a big issue with in-ear mic because certain headphones are very sensitive to placement on the head... much more so compared to a flat-plate coupler that generally gives consistent results everywhere. Not sure how it's been to you, but some headphones, like, say... HD800, generally run into this problem. Push the headphone backward such that your ears are more forward in the ear cups and the response will change completely. Now, granted, in general, this means in-ear mic will no doubt show most of what you hear when the headphone fits on your head, but at the same time, the measurements will only apply to you, and not to anyone else. In which case, it's not very useful as a metric. This is one reason why I haven't posted any measurement lately.

3) >5KHz region is arguably very important because that region dictates at least half of what you hear, so when your measurement method is only applicable <=5KHz, it's obviously not very useful, but that could just be me. Trying to "merge" results from two entirely different measurement methods is a very convoluted and nonsensical process, so I don't think that applies either. If you honestly want something consistent, perhaps try to apply compensation to the method that's more consistent >5KHz because in general, I think >5KHz gets progressively harder to measure compared to anything below.

4) Measurements are in general useless... but they do have some merits in making comparisons... provided the results are consistent and testable/verifiable. Actually, consistency is not even that big an issue, but verifiability is a very important part. It is for that reason that if you measure something that doesn't even look remotely like how at least someone else hears it, then some red flag will be raised and you'd be asked to verify it. If your verification process involves... well, yourself, then imagine how science would have worked if all experiments could only be proven by one person, who will eventually die/pass away. So long story short, you should use a method of measurement that can be replicated by someone else, in order for your work to be verified... and that will make life easier for everybody all around. It is for that reason that some of us started to post more info about our measurement systems... because we wanted to at least verify that our results will be consistent with the next guy.

5) On that note, Hands' results generally look more like how some folks would hear something, and that's why his method is verifiable. Yours don't really look "like anything", if you know what I mean. Not to nitpick but for instance, your HD800 measurements look pretty consistent, but your HD800X measurement doesn't look like anything else that anyone has ever measured, and that kinda makes your method somewhat "mysterious". We seek "verification", once again, and that's why we're asking you all these questions. But if you can only say "I did secret shit", then...

6) I think I went over this with you in length, but... do consider what you're measuring for. What do you want to get out of the measurements? Is it comparison against other headphones that other folks measured? In which case, you've gotta be a bit more "consistent" and use something that someone else can verify, otherwise it's going to be hard to correlate between your measurements and someone else's. If it's not for comparison purposes, but that you wanted to just know how something sounds like to you, then... dude, here's a quick way to test: listen to a sine wave sweep, and see whether you perceive the dips/peaks as measured, or if it's a straight and smooth road. That should tell you enough, and there wouldn't be a need to measure... or argue with anyone else about anything.

Personally, I measure now mostly to correlate between things I've done to mod the sound of a headphone, and whether it's consistent with what I hear... and that's that. No need to compare because my modded headphones already sound completely different from stock, and comparing against something else is just moot. So that's the other reason I don't post measurements.

But really, what do you wanna get out of measurements?

 

Catching up on some of the threads that I wanted to reply to now. This one's first.

@Bill-P
1) I mainly used the "air" term because that's what other people like to use. In other words, The HD800 subjectively had not only the most treble energy, but to me also the smoothest. The coupler measurements correlate best with this. The HD800 is super smooth on from 5kHz. Using a smaller mic capsule (WM61A) actually results in an even smoother FR with no resonance until 38kHz. Very linear from 5kHz to 20kHz, on par with good tweeters I'd say. I don't entirely agree that any emphasis in this region sounds nasty, because I think it depends on the nature. For example I have no issues with the Lyra Delos MC cart rising FR (10db at 20kHz I think), while even less emphasized treble, specifically around the air region, can make it sound very piercing with digital. However this does come down to system matching, cables and obviously also the transducers. Generally I would disagree that the FR is a good indicator if something will sound piercing here or not. I hear my HD800 as less annoying and colored than metal tweeters btw.

2) In this case I think it's wrong to think that higher consistency means higher accuracy, at least how you're describing it. Headphones do vary in sound depending on where you put them on your head and the in-ear measurements describe these variations accurately to me. The microphone insertion depth will change the results a lot and this does hurt consistency.
I also don't think that the in-ear measurements "only" apply to one person and not anyone else. I'm going to share the measurements that I took of @sphinxvc and @SteelCannon in July: https://www.dropbox.com/sh/s335paj8sc6zogv/AACzUXRpLLUAiCk21ZlBrCi6a?dl=0
These were very quick measurements and accuracy could be much improved if I had more time. Especially sphinxvc's HD800 measurements aren't accurate in the bass (mainly L channel). While I'm not going to argue that the differences are insignificant, they were generally pretty close up to around 4kHz, from where on the different ear resonances will make them look different. I would say the differences (<4kHz) are mostly within the general positioning differences you get with in-ear measurements. The most important thing that it showed me was that while all the 3 people in the folder showed much more 6kHz in their measurements than me, the difference was consistent for the HD800 and HD650, so I would assume that we generally hear the differences between two headphones as the same. This means to me that once you find out how one person's measurements correlate to your own impressions (i.e. Hands' measurements), you can probably gain more about the sound of another headphone than from coupler measurements.

3) It's not completely useless above 5kHz, just that you will get resonances that aren't there. I'd argue the same thing happens with the creatology foam coupler to a lesser degree. I think that it's impossible for one method to be accurate and "anechoic" across the whole frequency spectrum. The main takeaway from before was that you can't just take one compensation and apply it to the coupler measurements so they look like the in-ear results. Different headphone types will require different kinds of compensation. But I somewhat agree that it's better to settle on one thing that gets reasonably close at being relatively accurate across the band rather than trying to get "absolute accuracy" when that obviously can't exist. I was mainly trying to give my own perspective on why flat plate coupler measurements are inherently flawed and back that up with some measurements and (what I thought to be) a neat way to present the data.

4) You brought up a very important aspect here. Even then it seems like everyone on SBAF is still using their own couplers.

5) Personally I feel my own HD800X measurements are accurate to my ears . I haven't really changed the mods since I described them here:
http://www.superbestaudiofriends.or...superdupont-resonator.1147/page-30#post-57753
On the other hand I'd bet that with more data my measurements will start to look less "mysterious". Especially phones that Hands measured such as the Ether and Dharma. For now I'd argue that my measurements are no less verifiable than Hands's.

6) What am I measuring for? I'd say for an easier way to represent how things sound to me. Or, like you said, to help with modding headphones. This is actually the reason I tried the in-ear measurement method. Because the changes in the measurements weren't consistent with the changes that I heard with the HD800 mods. Well, that and curiosity. I also sometimes do experiments to show people who haven't taken measurements themselves how graphs can and will always lie. I consider the other thread such a "WIP" experiment. This is something that becomes very obvious when you take measurements yourself, but is hard to see when you don't know how the measurements were created and which compromises have to be made when measuring.
I also want to make up my own mind on headphone neutrality. For this I've been listening to speakers, comparing them against my HD800 and comparing various measurements for some time now. I find my in-ear measurements to be the most comparable to a listening position omni mic measurement calibrated with the B&K target. In other words a relatively flat FR with the headphone in-ear results tends to sound like the B&K speaker target at the listening position.
I feel you do need a more "absolute" reference than simply the stock version of a headphone you're modding, because what would be the target otherwise? The target obviously is "good sound", but how do you define that? How does that measure? Those are also things I'm trying to find out.

I also find the sine sweep method to be pretty inaccurate. In this sense I agree with @ultrabike. Sure, it will tell you about gross problems but hearing 0.5db changes in sine sweeps is a lot harder than simply listening to music. I also think that while a flat perceived sine sweep actually sounds pretty good with headphones, the same cannot be said for speakers. And at the same time I think that when you try to EQ speakers and headphones to have the same perceived FR with sine sweeps, they will sound very different with music. FWIW I do generally perceive the sine sweeps similar to the in-ear measurements, with the exception that a small bump around 5kHz in the graphs (like with the HD600 and HD800) probably sounds smoother with the sweeps. I also hear a dip around 6kHz with my HD600 both with music and sine sweeps, unlike my HD800X which I find to be flat, much like my measurements say. I don't know about you but I think this may simply be a case where we "hear differently".


@ultrabike
I feel the same way. A dip after 10kHz is also not a good thing at all.