DFHRTFs used at Headphones.com

Howdy all - prompted by a Twitter dialogue, I’m posting the HRTFs we’re presently using for compensation of headphone measurements, and their provenance.

@taronlissimore will probably tell me I’m silly for my formatting later, and that’s fair, but the only way I could figure out was linking google sheets.

4128 DF HRTF

This one is based on the kludgiest data we have currently. This DF HRTF was derived from the free field data that Brüel & Kjær make available for the 4128C with the straight canal pinnae, but it is limited in spatial breadth/resolution—it would ideally have more underlying data at a finer spatial resolution at the sides and rear, but B&K dies not provide this data with the 4128. As you can probably see, it’s been smoothed a fair bit - variably by frequency - because the “free field” that B&K used for the 4128 wasn’t all that free, and there were meaningful modal resonances in the response. In the long term, this is going to be the most likely next target for replacement as a result, so watch this space.

5128 DF HRTF

This one also comes from the official data from B&K, which thankfully is a lot more actually free field this time. This is the HRTF we’re happiest with at the moment, and while we’d of course jump at the opportunity to get even better data down the line, this is the least likely to be revised in the near future. Both the 5128 and 4128 averaging math was done by the eternally lovely Oratory1990 in correspondence with me, so big shout out to him here! We’ve done a little bit of manual smoothing to get rid of rockiness in the upper midrange and lower treble, because we think the compensation baseline for judging transducer-based devices should be reasonably smooth, as no one intentionally tunes a ‘rough’ response in their headphones, and leaving small wiggles in the response might make people assign undue importance to these features.

KEMAR DF HRTF using the KB500x pinnae

This is the one we’ve most recently replaced as part of a group effort between Oratory1990, myself, and @listener. Oratory1990 took the KEMAR with the KB500x pinnae attached to an HRTF measurement lab where we were able to get a full host of individual HRTFs for the head, and then he supplied @listener and I with the resulting data. @listener calculated the Diffuse Field HRTF from the data, and then I did some post-processing to get a result that made sense because, as you can see from the video here, WARNING: LOUD, the lab isn’t perfectly anechoic, so windowing was necessary to remove the influence of reflections below 2 kHz or so. After all was said and done, we got a DF HRTF we are very happy with—much happier than when we were using the DF HRTF for the wrong pinna(!) for our data using this ear.

KEMAR DF HRTF using the KB0065 pinnae

This is our newest HRTF and probably the least often used since only @listener uses this set of ears. It comes from the SONICOM HRTF dataset and was again calculated from constituent Free Field HRTFs by @listener. The measurement methodology can be found in part here, though there were a few things that had to be confirmed in private communications with the people responsible for measuring (many thanks to Isaac Engel, Lorenzo Picinali, and Ollie Turvey for assistance). In the SONICOM dataset, DF HRTFs for the KB0065 and KB0066 + GRAS RA0045 occluded ear simulator are measured, but we only use the DF HRTF of the former KB0065 pinnae because the KB0065 and KB0066 pinnae are actually asymmetrical, the clone rigs used by most people (eg. @listener, @MRS, and GadgetryTech) are actually solely based on the KB0065 right ear and an identical (but mirrored) copy for the left ear. This is another one we’re quite happy with.

4620 + Human DF HRTF (JM-1) [for use with IEMs measured on 5128]

@listener has a full post explaining this here, but the gist is that our pal @Joel did a mathematical estimation based on the human DF HRTF from ISO 11904 of what a DF HRTF using the 5128’s ear canal but human outer ears would look like, and our initial measurements seem to confirm his math. We’re still confirming our measurements of the 5128’s ear canal transfer function, but once we have we may update this—but if the existing JM-1 baseline is close enough, we also may not, so watch this space for updates

@listener has a mocked up “delta target” based on an average of the differences of the type 4.3 and 60318-4 couplers - this methodology in inherently and always inaccurate for any specific case due to the variation in output Z, but if you would like to see it anyway, it is here: ∆ 4620 + Human DF HRTF (JM-1) [for use with IEMs measured on IEC 60318-4 Couplers]

His comment is “This is easily the HRTF we are least happy with, but it has unfortunately become a bit of a methodological necessity if we’re going to use the same bounds for headphones as we use for IEMs. This is based on the JM-1 DF HRTF above, but altered with a delta calculation based on data from Oratory1990, csglinux, and Sean Olive measuring the difference between IEMs measured on IEC 60318-4 couplers vs. the 5128. We encourage people to be skeptical of results published with this baseline, or really any measurements of IEMs taken on IEC 60318-4 couplers.”

Credit to @listener and @oratory1990 for the majority of these HRTF calculations, I am but the messenger.

3 Likes

Just one thing to note - these are not target curves the way you might see with preference elements from say Harman or any kind of tilt applied. They contain no preference data.

2 Likes

Yeah, EQing to straight DF…well, there’s probably like 5% of people who would prefer that, but odds are that you, the reader, are not one of them. DFHRTFs are just that: the head related transfer functions of these mannequins in a diffuse field. Olive 2013 showed that DF is roughly equivalent to the harman in-room baseline measurement, so summing these with the shelf filters used in Harman, or applying a constant slope, or really anything else you’d think would be suitable for a speaker “room curve” is the play here.

We’ll be making a tool available in the future that should take some of the work out of doing just that for folks, in fact…

3 Likes

I think that this should be mentioned more often. People are used to seeing measurements compensated to Harman where flat might be good. I tried EQing to DF-1dB/octave when I first saw 5128 measurements of the headphones I own because I didn’t know any better. I used the preset for some time before I found out I don’t really like it. This makes me afraid that people thinking that flat is good will start a trend of manufacturers making headphones tuned to DF-1dB/octave to have “well measuring” headphones people want.

I personally prefer less ear gain compared to DF and the elevation to be shifted by a little bit to higher frequencies. I don’t mind another elevation after 5K. Moreover, I like warmer headphones so I prefer a small elevation in the midrange.

Yeah and this is where preference windows come in. For people who really want to dial it in, you’d want to apply whatever tilt you prefer. So for example, if you like a warmer sound signature, go with 12 or 14dB to start. These are just the compensations for each measurement rig, which has no preferences.

1 Like

It should be noted that linear slopes aren’t the only way - again, it’s best to view DF as a baseline for applying a preferential EQ, which might well include notch or peak filters in addition to shelves.

2 Likes

Thanks, Mad_Economist. Looking forward to the KEMAR DFHRTF with KB500X pinnae! Are the loudspeakers linearity, room, and methodology compatible with the ones used for the 5128 DFHRTF or it doesn’t really matter?

1 Like

Indeed, and that’s also MUCH easier to do when compensated.

Just to expand a bit here on what @Resolve said, this is actually about 40% of why I’m ardent that DF is the way. A diffuse field is a standardized sound field (where sound is non-incident/power is evenly distributed across space), just like a free field. A properly conducted free or diffuse field measurement is compensated for the response of the speaker(s) used, and has no room or other unique interactions. In premise, four labs on four continents, using four different speakers, should be able to get strongly matching results with the same head as long as each uses the correct methodology.

3 Likes

Sorry, where can I find the official set of HRTFs with free field compensation?
Is there even some SOFA files maybe?