• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

My HRTF Target vs Harman Target

neutralguy

Member
Joined
Apr 16, 2019
Messages
65
Likes
127
In a previous post, I detailed how I measured my own Harman like target for my own head, which involved measuring my own HRTF using a loudspeaker and an in-ear microphone. Even headphones that follow the Harman curve on a dummy head measurement needed correction on my head. Is it due to my head's deviation from average, or something else? I investigate in this post.

To summarize my previous post, I measured and computed my own HRTF by subtracting the free air response of an in-ear microphone (MS-TFB-2) from the response when mounted in my ear, as my speaker (Revel Salon2) played REW's sine sweep and I sat in the sweet spot. Harman showed that users prefer headphones to have a response that's basically an ideal speaker's in-room response transformed by the HRTF. When Harman did this with their speakers and headphone measurement rig (Gras 45 with mannequin), the result was the Harman headphone target. Applying the same method, an ideal headphone for me should have a response that is the sum of my HRTF and this Ideal loudspeaker target. Let's call this target response the Neutralguy Target:

loudspeaker headphone target.png


How does this compare to Harman's target? The two are not directly comparable because my measurement was made at the entrance (BC) of the ear canal, mostly blocked by the measurement mic. Harman's was made at the dummy's eardrum location. Fortunately, the relationship between these measurement locations has been well studied. In Figure 4 of Denk 2018, we can see HRTF measured at blocked ear canal entrance and eardrum averaged over a number of individuals. I converted their publically available data into REW, and computed the difference between the two as a transformation between them. It turns out that this transformation is similar for both free field and diffuse field, so it seems applicable to our in-room sound field, which is somewhere in between.

Here is how my target compares to Harman's, with the eardrum transformation applied. The original BC target is also shown, offset lower, for reference.

neutralguy vs harman target.png


We see that Neutralguy Target shows a broad bump in the 200-1khz region, and then a bigger bump around 2-4khz. Since we derived the targets using the same procedure, the difference comes down to our measured HRTF's. You can see Harman's HRTF in Figure 5 of Olive 2012, which does not show these bumps. Let's further investigate this discrepancy.

Headphone Response

I measured a Sennheiser HD650 headphone, which in amir's measurements using a Gras 45CA shows good compliance with the Harman target. I've applied the eardrum transformation to my measurement to yield the HD650 eardrum trace for comparison with both targets.

hd650 vs targets.png


We see that the HD650 shows good agreement with Harman target from about 1kh to 8khz, as in amir's measurements. Above 8khz headphone measurements are known to be highly variable, and in the bass it varies by earcup seal.

The headphone's measured response is heavily shaped by the ear in the high frequencies. To show this, I simulated a head without ears by covering a pillow with a silicone sheet, and mounted the headphones on the pillow. The in-ear mic was placed flat against the pillow inside the earcup. Another comparison measurement is the headphone's free space response, taken by dangling the microphone just outside the unmounted earcup. We compare this with the HD650's normal response when mounted on my head. Each response was averaged over at least 3 reseatings:

hd650 response.png


Compared to free mount, we see that there's a gain below 2khz whether mounted on the pillow or my head. It's caused by the headphone seal creating an enclosed volume. Below 200hz the gain varies depending on tightness of fit, and the pillow's compliance may have allowed a tighter fit.

Above 2khz it's apparent that my ear plays a critical role in shaping the HD650's frequency response, and we saw that our response matches the Gras quite well in that region. This suggests that my ear and cheek are comparable to the Gras 45CA's as far as shaping sound magnitude, at least when sound comes from a headphone's near field 90 degree direction. Therefore, differences between my HRTF and their HRTF is likely not due to the ear, but the rest of the body.

Body RTF

To isolate contributions to HRTF from the ear and the rest of the (head and) body, I computed a version of my HRTF that aims to remove contributions from my ear by simulating my head with no ear. I measured HRTF as before using my loudspeakers, but with a sheet of silicone between my ears and the mic. The silicone keeps the mic from being placed inside the pinna, and provides some isolation from its cavity. The resulting measurement is the Body RTF. The regular HRTF is plotted for comparison, and their computed difference is imputed Ear RTF.

BRTF and ERTF.png


We can see that above about 2khz, the ear contributes more, while it's all body contribution below 1khz. The Body RTF's contribution is enough to explain Neutralguy target's bump above Harman between 200-800hz, but the target difference in the 1-2khz region is higher than the Body HRTF. It could be due to measurement error of this crude approximation, or our ears may be transforming sound differently when sound comes from loudspeakers, even if they appear similar coming from a headphone. I would guess that the rise around 200hz, given the long wavelengths involved, necessitates the presence of my entire body to create.

As a secondary check (not shown), I also measured using a slightly different setup, where to avoid the ear's contribution I simply dangled the mic above the ear and below the ear and averaged the two measurements. The result looks bumpier but is in broad agreement with this covered ear result.

This result offers some confirmation that differences between my target and Harman's target is due mainly to my body versus their dummy's. Sound from a headphone (HD650 at least) arrives to our eardrums with a similar response, but not sound from loudspeakers, which interacts with the body.

Neutralguy vs Harman

So neutralguy, you're saying your target might be better than Harman's, even for the general population? But didn't Harman show that the average user prefers their target? Well yes, but their adjustment possibilities were limited to broad bass and treble adjustments, above and below 200hz (Figure 7 of Olive 2012). There was no option, say, to boost 500-1000hz, so we can't rule out that such an adjustment would be even more preferred.

I should point out that one of the main deviations between my target and Harman's is in the 2-4khz region. This is the region where we may be adapted to a depression in the response from loudspeakers. One cause is the stereo interference dip of about 4db around 2khz, as shown in Figure 7.2 of Sound Reproduction. Additionally, this is the region where many loudspeakers cross over from the midrange to the tweeter due to a constriction of the midrange's dispersion, which when not complemented perfectly by the tweeter results in a dip in the in-room response (examples measured on this site: Genelec S360, Revel F226Be). This means that we may already be accustomed to a depression in this region based on loudspeaker sound, so a headphone that "fixes" this may not even be preferred, even if it's a more technically ideal response.

If you have a set of headphones tuned to Harman, you can tune them most of the way to the Neutralguy target by adding these eq settings:

+1.5db 600hz Q=1.2
+5db 2400hz Q=2

harman to neutralguy eq.png


These settings are intentionally conservative. They bring you only part way to the Neutralguy target, in case there are any peculiarities of my ears and body that do not match yours, or you are accustomed to Harman's target for reasons above. I've left 7khz+ alone. There's a definite rise above 10khz in my HRTF and in the free field HRTFs in Denk 2018, but not in their diffuse field curves. I do think a rise is appropriate, but that's also a region with high variability from exact headphone positioning.

TLDR

I compared my measured HRTF and resulting headphone target against Harman's. The areas where we differ appear to be due to my body versus their mannequin. I would be interested to see how others like the eq settings above that transform the Harman target into mine.
 
Last edited:

Feelas

Senior Member
Joined
Nov 20, 2020
Messages
376
Likes
307
Just wondering - shouldn't you also count in room reflections (e.g. via averaging the HRTFs), since pinna boost is direction variable? The resultant HRTF is not the same for different angles of attack. In other words, HRTF is a family of graphs, not merely one HRTF. Averaging *might* somehow explain how you can arrive at a different curve if not taking all of the data points at once. I don't remember (but am pretty sure it did) whether the original speaker study measured only the direct sound on GRAS or whether it was averaged from numerous measurements in differing angles, arriving at an averaged HRTF w/o deg dependence.

Interesting idea nonetheless.
Harman showed that users prefer headphones to have a response that's basically an ideal speaker's in-room response transformed by the HRTF.
Just wondering: wasn't the "preference curve" more of a "how the customer preference for headphones relates to in-room reference" instead of "preference is the in-room response transformed thru HRTF"? Very similar, but slightly different in an important way.
 
OP
N

neutralguy

Member
Joined
Apr 16, 2019
Messages
65
Likes
127
Just wondering - shouldn't you also count in room reflections (e.g. via averaging the HRTFs), since pinna boost is direction variable? The resultant HRTF is not the same for different angles of attack. In other words, HRTF is a family of graphs, not merely one HRTF.

Yes, the HRTF is direction dependent, which is why the HRTF is specified with respect to the sound field, the combination of sound directions. The sound field Harman proposed is that of a loudspeaker in room, which includes direct sound from loudspeaker plus reflections included already.

On a related note, it's confusing exactly which speaker setup Harman used. Section 2.3 of Olive 2012 says they averaged the HRTF over their 7.1 system. It's possible that Harman's HRTF shows lower gains because they used the 7 speaker average. However, in Olive 2013 they used a stereo system, and first equalized the headphone to the in-room response of "the" loudspeaker (Figure 5).

Just wondering: wasn't the "preference curve" more of a "how the customer preference for headphones relates to in-room reference" instead of "preference is the in-room response transformed thru HRTF"? Very similar, but slightly different in an important way.

They experimentally determined the two to be similar, so I'm just treating them as the same in this post to focus on the HRTF differences, if that's what you're referring to.
 
OP
N

neutralguy

Member
Joined
Apr 16, 2019
Messages
65
Likes
127
Looking more into the confusion in the Harman papers. The first publication that I know of the Harman curve is in Olive 2013, where in no uncertain terms they used a stereo system and a Gras 43CA, and "equalized [the headphone] to match the in-room target response of the loudspeaker measured at the DRP". But the Gras 43CA has a cylinder for a head and a pole for a body, and they got a measurement that looks similar to their earlier mannequin measurements? Loudspeaker on the same side as the ear shows an HRTF bump in my measurements around 10khz preceded by a depression around 7khz. Similar patterns can be seen in the literature [Denk 2018, Moller 1995] for free field frontal incidence. Theirs in Figure 5 shows no such bump pattern, more consistent with spherical diffuse field curves that include above and below directions.

Let's say it was a mistake in the description, and what they actually did was equalize to the response derived from their earlier Olive 2012 paper, where they used a mannequin and averaged their 7.1 system to derive a free field like curve. I don't have a 7.1 system, but I did something I think should be comparable, where I rotated my body in each of 7 directions, mimicking the angle of the 7 speakers relative to body. While some of the differences between mine and Harman's curve are reduced, a +16db bump at 10khz remains.

neutralguy 7.1 target.png


It's conceivable that Harman may have made adjustments not mentioned. Removing the bump at 10khz for a headphone is a reasonable hedge against an ear piercing peak, given inconsistencies in high frequency response due to headphone seating variance. At the end of the day, a well grounded derivation doesn't guarantee a target will be preferred, and their subjective testing does show users prefer theirs relative to the some alternatives. Still, it would help to have a replicable result that others can build on when investigating potential improvements.
 

jae

Addicted to Fun and Learning
Joined
Dec 2, 2019
Messages
915
Likes
1,075
Just wondering - shouldn't you also count in room reflections (e.g. via averaging the HRTFs), since pinna boost is direction variable? The resultant HRTF is not the same for different angles of attack. In other words, HRTF is a family of graphs, not merely one HRTF. Averaging *might* somehow explain how you can arrive at a different curve if not taking all of the data points at once. I don't remember (but am pretty sure it did) whether the original speaker study measured only the direct sound on GRAS or whether it was averaged from numerous measurements in differing angles, arriving at an averaged HRTF w/o deg dependence.

Interesting idea nonetheless.

Just wondering: wasn't the "preference curve" more of a "how the customer preference for headphones relates to in-room reference" instead of "preference is the in-room response transformed thru HRTF"? Very similar, but slightly different in an important way.
HRTF is for one position/incidence of soundfield, which in the case of this research is the frontal position from a sound source (in that case, a typical equalateral stereo setup in a semi-reflective room)
 
Top Bottom