[JMS]

In the thread in this forum on the AES2012 paper about perception and measurement of headphone sound quality, I asked the Harman research team whether they have considered measuring desired headphone frequency response using loudspeaker playback on a dummy head as reference, similar to what is done by the Smyth Research Realiser A8.
 
Well, I've taken up the task myself, using binaural microphones in my own head, measuring in-room using Room EQ Wizard 5.0. The microphones used are the in-ear binaural microphones Sound Professionals MS-TFB-2, which when mounted, allows over-ear headphones to be worn as usual. Loudspeakers used are Revel Salon 2 in a smallish room. Note that neither the frequency response of the microphones nor loudspeakers matter, since we will only care about relative differences.
 
To be precise, the measurements are as follows:
 
a) Measure in-room response of loudspeakers, using binaural microphones mounted on stand.
b) Measure in-room response of loudspeakers, using binaural microphones mounted in ear.
c) Measure response of headphones when worn over binaural microphones mounted in ear.
 
- Subtracting (a)-(b) gets us the HRTF of my head in-room, at least in terms of frequency response.
- If our speakers and room were both ideal, then we would like our headphones to have a response such that (c) is equal to (b).
 
For spatial averaging to minimize comb filtering effects, I took measurements of (a) and (b) in each of 3 locations: center, and 15 degrees to the left and right. Okay okay, I didn't really measure 15 degrees exactly.
 
Here are the 3 curves for (a), after 1/6-octave smoothing:
 

 
 
Similarly, here are the 3 curves for (b):
 

 
 
The REW software doesn't support averaging the curves, so here are the right15 curves superimposed on each other:
 

 
Red curve is free standing, green curve is in-ear.
 
Continued next post.

 

[JMS]

My personal in-room HRTF frequency response is implied by the difference between the red curve (free standing) and green curve (in ear). Given that these two aren't spatially averaged curves, I wouldn't look too closely at small peaks and dips of low bandwidth. Looking at wholesale differences though, there does not appear to be any distinct peaks and dips of significance as in various standard HRTF curves. The only major difference I note is a broad boost of 2-5db from 600hz to 15khz. What this is attributed to and whether it is real or an artifact of my technique, I don't know.
 
Moving on to (c), here are the responses of the Sennheiser HD650 (red) and Koss KSC75 (blue).
 

 
I'm surprised how flat the HD650 is. The HD650 curve looks quite different in the upper frequencies than either the headphone.com or innerfidelity.com graphs, so I'd like to do more testing and measurements before drawing strong conclusions from it. I briefly played with a parametric eq to "fix" the HD650's depression at 7khz and the KSC75's broad midrange depression from 130hz to 3khz, and the tonal balance did seem improved in both cases.
 
In the coming days I will be repeating this measurement for Sennheiser HD700 and AKG K550.
 
In the meanwhile, I'd appreciate comments from the various measurement experts on this forum. I'm sure there are some things I missed or could improve upon. Are my techniques valid, and are the results consistent with what we know?

 

[xnor]

Quote:

For spatial averaging to minimize comb filtering effects, I took measurements of (a) and (b) in each of 3 locations: center, and 15 degrees to the left and right. Okay okay, I didn't really measure 15 degrees exactly.

What does that mean?
 
The speakers should be located in an equilateral triangle (equal distance from the head to the speakers and also between the speakers), so at +- 30° pointing to the center of the (dummy) head. For averaging you could take the left speaker & left ear measurement, right speaker & right ear measurement and you could even rotate the head slightly to the left/right. These six measurements should be enough for averaging.

 

[JMS]

Quote:

What does that mean?
 
The speakers should be located in an equilateral triangle (equal distance from the head to the speakers and also between the speakers), so at +- 30° pointing to the center of the (dummy) head. For averaging you could take the left speaker & left ear measurement, right speaker & right ear measurement and you could even rotate the head slightly to the left/right. These six measurements should be enough for averaging.

 
It is to minimize the effects of comb filtering from reflections, diffraction, or other artifacts from the loudspeakers or the room. It is a standard technique used when measuring loudspeakers, such as in Stereophile's measurements.

 

[xnor]

So if I understand it correctly, in case a) you put the speaker in front of the mic (where the mic is pointing to)  and in case of b) you put the speaker in front of your head (where you look at if you look straight ahead) and in both cases also at about +- 15° (left/right)?
For b) you really should use a 30° angle (and maybe +-5° for averaging?). After all you want a HRTF that is similar to your normal listening situation (speakers and your head set up in an equilateral triangle).
 
What was the distance between mic/head and speaker?

 

[ultrabike]

 

[mikeaj]

I'm going to join, if you don't mind.
 

  

 
 
 
Actually, there doesn't seem to be a whole lot of dependency on angle, at least with the (1/6th octave smoothed) data there.  I wonder how much energy is coming off of the walls in this small room.  If the angle of incidence is off, whether by an "incorrect" speaker position or some sound waves not coming from the direct path, then it seems like the transfer function would be thrown off, and not necessarily in a way that could be compensated for by averaging across positions or different tests.  That's just my guess, though.

 

[xnor]

Quote:

Actually, there doesn't seem to be a whole lot of dependency on angle, at least with the (1/6th octave smoothed) data there.

HRTFs are highly dependent on angle. The changes in the frequency response can be huge if you change the angle by a few degrees.

 

[liamstrain]

I'm not sure you can really factor HRTF from the microphone placement - unless those mics are sitting against your eardrums, and do not affect the physical space of the ear canal... 
 
 
But interesting results. Curious to see what repeated testing and additional variation shows. 

 

[Puranti]

I'm sorry if didn't understood correctly but, what is the point of these measurements ?

 

[ultrabike]

Quote:

I'm sorry if didn't understood correctly but, what is the point of these measurements ?

If I understood the OP correctly: To qualify and quantify the effect that "using loudspeaker playback on a dummy head as reference" has on "measurement of headphone sound quality" in terms of "headphone frequency response". And to correlate those measurements to "perception" "of headphone sound quality."
 
BTW, your measurements on the KSC-75 seem to correlate well with what I hear: Some upper bass emphasis, a bit bright, and somewhat sunken mids. To me your measurements look similar to these ones:

 
Your HD650 seems to exhibit a deep notch at around 7kHz. Here are some other results showing a similar gradual notch starting at about 5kHz, reaching a minimum at around 7kHz and somewhat recovering at around 10kHz (somewhat similar to your results - Differences might be due to mic placement and compensation of course):

 
Tonal balance described above matches well with your measurements and indeed looks fairly flat up to about 5kHz. There might be some product variation in regards to the HD650. Here is a second HD650 measured using a similar procedure as the one above but with different results:

 
If anything, your measurements seem to roll low and mid bass regions a little faster relatively speaking. Thanks for sharing your results!

 

[JMS]

Ah, it's apparent that I need more clarification. The loudspeakers are set up in my living room for normal listening, forming an equilateral triangle with the listening seat about 7 feet on each side. The measurements from the "center" position are taken from the position of my listening seat. The "left" and "right" measurements are taken with the seat moved about 1.5 feet to the left and right respectively.
 
a) For measurements (a), I placed the microphone on a stand resting on my listening seat, oriented roughly the same way they would in my ear.
b) For measurements (b), I placed the microphone in my ear, while I sat as I would normally, face forward, in my listening seat.
 
 
The point is to see what headphone frequency response would mimic the same frequency response my ears would hear from loudspeakers in a room.

 

[ultrabike]

So you are attempting to null speaker - room interactions by using measurements in a) and b). Cool!

 

[JMS]

Thanks for posting these comparison graphs, ultrabike. Do you have links to the original descriptions? I'd like to know whether smoothing was applied to them. My headphone graphs are unsmoothed, and look more like the ones you posted after 1/6 or 1/3 octave smoothing.

 

[ultrabike]

I think they were smoothed using 1/6 octave.
 
and yes, the smoothing may explain some of the measurement differences. Your results look remarkably similar though, and again, correlate well with what I've heard.