3rd party plots

home
back to Learn
back to How to interpret graphs

post separation

where to find 3rd party frequency plots

For most headphones you can find various types of graphs. Each site has their own measuring method, their own preferred ‘compensation‘, their own scales and their own types of graphs. This makes comparing graphs from different sites very difficult.

The best correlating measurements (IMO) to actual headphones that actually say something about the perceived sound are those from Rtings. This website is now doing headphone measurements as well and is looking quite promising.
Due to the HATS (and compensation used) the treble may show large peaks and dips that in reality may not be that bad. Also bass response seems to roll-off sooner and steeper than most other measurements.

The English Golden Ears  site seems abandoned after July 2014, the Korean Golden ears site  has been active till June 2016. After that date no new measurements had been uploaded. Too bad as this is/was a quite valuable site.

Also Innerfidelity has very useful graphs but alas slightly ‘wrong’ compensation was chosen from the start and everything above 1kHz is not represented as it is perceived in reality.
When you overlay the plot below over Tyll’s plots and regard the green line as ‘neutral sound’ then these plots will make a lot more sense. More on this further down this article.

For HeadRoom I recommend to use the ‘Frequency Response Compensated’ option in the graph typebox.

Changstar which has now reinvented itself as Superbestaudiofriends also has frequency plots but the older ones (the ones without the Harmonics below it) lack precision below 200Hz and around 15kHz the measurements seem to be a bit (3 to 7dB) pessimistic.
Some of the plots there (by others than Purrin=Marvey=Marvin) and Ultrabike) are quite inaccurate though. The added value Changstar has is the CSD (waterfall plots) and their THD (Harmonic distortion) plots. The newer plots (those with the harmonics) are much better but also seem to display the treble area a tad too low.
Plots that are sloping according to the ‘B&K room compensation’ plot are actually ‘flat’ sounding.
So slightly elevated lows and slightly ‘lower’ than a horizontal line is about similar as a horizontal line in plots measured on this website.

This Russian site (Personal Audio.RU) also has usable plots but online-translation is almost obligatory if you want to read the article.

This Russian site has the same plots as Personal Audio.ru but has a few more plots and also some hires plots (from 10Hz to 45kHz). Compensation around 5kHz seems to be ‘off’ though.

Another site that has some useful information and nice graphs but is focused somewhat more on in-ear monitors is Rin Choi’s blog. This site is abandoned though.

For all these websites one thing is certain. For comparing headphones with other headphones on the SAME website they are perfectly suited.
When you want to find out if the plots are ‘reliable’ then try to find plots of headphone(s) you own or know well or ones that are are well documented such as the Sennheiser HD600 and HD650.
When the plots of known headphones match your personal experience or look very much like ‘known’ measurements chances are you can trust other measurements of the same person/site as well (no guarantee though)

Graphs/ plots differences

Not all graphs and plots found on the web are the same. This is because of the differences in measurement techniques, compensation (for the microphone) applied, used ‘dummy head’, position of the driver, seal of the pads, ideas/beliefs of the one creating the plots, variations in manufacturing of the headphone (drivers), used output resistance of the amplifier driving the headphones, the scales differ considerably seemingly making plots ‘flatter’ or ‘bumpier’ and the applied ‘smoothing’ can make plots differ considerably.

So when interpreting graphs you need to be aware of the frequency scale which should run from at least 20Hz to 20kHz. Also the vertical scales (dB’s) are different. Some are 5dB per division, others 10dB/division. The height of each division also differs from site to site, something to take into account. Both the vertical scale and horizontal scale are logarithmic. The vertical scale in dB’s appears to be linear BUT dB’s in nature are logarithmic just like our hearing is. Because hearing is very close to the dB scale it is used in audio.

Depending on HOW frequency plots are ‘compensated’ to a specific ‘sound field‘ depending on the views of the one publishing those plots and the way a living (listening) room ‘alters‘ the actual frequency range at the listening position some are of the opinion this must be incorporated in the plot. An example being OLDER plots Headroom, which is a great and very informative website.
The newer plots in ‘build a graph‘ section and selecting ‘Frequency Response Compensated‘ in ‘graph type‘ will yield very accurate plots.

For Innerfidelity plots a ‘correction’ needs to be used in order for the, extremely valuable, plots to make ‘sense’ in an acoustic representative way. Everything below 1kHz is correct, however, above 1kHz a downwards sloping line should be drawn in ending at around -12dB at 20kHz.

comp curve IF

This compensation should be superimposed over innerfidelity plots where the 1kHz points match in level (may not be at 0dB level but could be several dB’s lower in some plots).
Below an example of how this can be done. I used two plots of headphones that are known to be close to neutral sounding HD650 and ETHER. When you regard the green line as an actual ‘horizontal’ line (considered audible flat when completely horizontal) the plots have a higher correlation to how they are perceived.
Above 6kHz the peaks and valleys in these plots are partially caused by the used Pinna and ear canal in the used dummy head.
The upper plot is the original innerfidelity frequency response from 1 of the pdf’s with the compensation line drawn in.
Below that plot the original innerfidelity pdf but with the 1kHz- 22kHz part ‘de-tilted’.
This is rather quickly done using MS paint so not the best possible effort.
I think the de-tilted plots resemble the way I hear these headphones a LOT better than the original plots ‘suggest’.
4 x Tyll compensated

About the seemingly always present 10kHz peak:
It looks like the 10kHz peak is a resonance in the HATS that is not compensated.
When there is no peak is present in a headphone you may still see a peak there (because of uncompensated resonance) unless, of course, the headphone itself has a ‘dip’ there.
When the headphone does have a peak around that frequency there will also be a peak visible, maybe slightly higher in amplitude or wider in shape.
This makes evaluation of the treble quite difficult (impossible ?) as the 10kHz MAY be there and it may NOT be in reality.

Headroom (Headphone.com) I can recommend when you use the ‘Frequency Response Compensated‘ graph option in the ‘graph type‘ box. Note that this is not the default ‘compensation’ so has to be set manually.
That setting will give a decent representation of how the headphone sounds.

On the OLD headroom website plots are over-compensated (a bit like those of IF). Below some ‘manually compensated plots’ of familiar headphones where I drew in a green line that shows the ‘room equalization correction’ that should not have been applied. If you tilt the hole plot so the green line is horizontal you can get a better idea how it actually sounds to ‘trained’ ears. This is shown in the plots below.
These plots are not used on headroom website but can still be found on various forums for instance.

Those familiar with the headphones shown below will agree they do not sound as shown in the upper row. I mean… a DT880 and K701 that have a substantial amount of of bass?
A D7000 that sounds quite dark, warm/bassy enough that it has 7dB more bass than treble ? .. no not really.
This is not how I perceive these headphones at all.
                                                 Click on the picture below to enlargecompensated headroom plotsThe plots below the ‘original’ headroom plots have been ‘de-tilted’ as it were. The room equalization that has been applied to these plots and thus has been removed. The green line that was slanted has been tilted to the left so it is ‘flat’.
The K701 now doesn’t seem to have a warm tilt as it appears to have in the plot above it, instead the mids are very balanced and real. There is a slight treble peak and the subbass is rolled off (really it is K701 fans !). The HD650 now doesn’t seem to be a boomy and severely rolled off headphone but suddenly seems quite balanced with a slight warmth hump (100Hz to 300Hz) and isn’t as rolled of as the original plots would indicate. It has only a very minor downwards slope in the treble. The DT880 on the bottom plot is exactly as I hear it. Very good mids, lean on the bass and a peak in the treble region (which gives it its fake hyper detail). The GS1000 still has very peaky treble and a firm bass boost in this plot. The D7000 is remarkably flat, only lacking in subbass extension and a tad brighter than the HD650 in the treble.

different measurement techniques

Another type of compensation is needed when measurement rigs are used when measuring microphones are placed inside tubes or other types of artificial ear canals.
Some people are of the opinion that such measurements are closer to ‘the truth’.
Such tubes ALTER the frequency response of the incoming sounds and to create a plot that has a high correlation to what is actually HEARD all the alterations the tube has made has to be undone again. This is not easy to do and requires digital filtering.
For that reason a flat-bed measurement rig (microphone flush to the baffle) is used so only the applied sound is measured. No compensation for higher frequencies is needed in this case.

The problem with most graphs out there is that it isn’t really clear HOW they were obtained and exactly WHAT ‘corrections’ have been applied. One of those ‘corrections’ is the HRTF (Head Related Transfer Function) which come in various forms. When a HATS (Head And Torso Simulator) is used the microphone is placed inside the head and is connected to the outside world via a tube and another tube simulates the Eustachian tube that ends up in the throat cavity. This tube is normally closed and opens when swallowing to equalize the pressure on both sides of the eardrum so the eardrum won’t bulge to one side.

An actual microphone, when held in free air, measures very flat.

When a microphone capsule is mounted on a plate it won’t measure flat any more either when used to measure S.P.L. (Sound Pressure Level) in free air, this also requires some compensation if used for that purpose when mounted in a larger surface.

Because of this measurement microphones have a small diameter ‘head’ often elongated and broadened away from the capsule to house parts and make it easier to clamp on a microphone stand or to hold.

The microphone inside that dummy head registers the sound differently though because of the tubes and those differences are substantial due to (broad spectrum) resonances around 3kHz. This ‘lifted’ part of the frequency spectrum needs to be compensated to obtain a close approximation of the sound pressure that is just outside the ear (entry hole). If an artificial head were used and the microphones were placed on the surface of the head this 3kHz peak does not have to be applied as there is no tube in front of the headphone that causes this peak.

This tube to the microphone is partly used when in ears are measured and as not the whole ‘ear canal’ is used one and that very small closed chamber that now exists is smaller and closed on both ends and needs a different compensation as when free-field (far-field) or near-field measurements are done.

Still not quite there though as the pinnae (the actual outside ear) also influences the incoming sound pressure and also needs to be compensated. Compensation for this is different when HATS is used for measuring speakers or registering music/sounds versus when it is used in a small closed compartment (a headphone).

 The pinnae placed on the outside of the head next to a microphone also creates small errors that need compensation when measuring free-field sound coming from the sides of that microphone.

Another compensation factor that needs to be included is the spherical sensitivity of a microphone which is not ‘flat’ for all frequencies. This is needed if such a head is used for far-field measurements (microphone far away from the sound source).

For headphones where the sound is directly coupled and not angled (or just slightly due to angled drivers) some correction factors do not apply. The coupling between the headphone and microphone is more direct (when a good seal is achieved) and thus yet again needs slightly different compensation.

The individuals making the measurements with the equipment at hand must make the right decision to use the appropriate corrections of the circumstances surrounding the microphone (H.A.T.S.) itself.

So compensation is NOT for what the ears hear. Shape of Pinnae and ear canal, are not compensated but the, in itself flat microphone, is fed an altered and frequency dependent S.P.L. that needs to be compensated to emulate what’s being fed into that ear canal from the outside.

It appears that sometimes incorrect compensation is used for the wrong reasons based on personal ideas or incorrect assumptions.

post separation

home
back to Learn
back to How to interpret graphs