Oratory1990 Canjam Presentation on Headphone Sound

[solderdude]

Preference can only be established under blind circumstances. Liking something is different. And the Harman target is not a fixed

I like many headphones but do have a clear preference. That preference is not established under blind circumstances.
The Harman curve is just an average which most people prefer.
Of course there is a substantial tolerance.

 

[Curvature]

I like many headphones but do have a clear preference. That preference is not established under blind circumstances.

I think if a person is really interested in "knowing themselves", they would acknowledge that there are aspects of themselves they have no control over, which are largely automatic, which rely on biology. That's the realm of preference.

If you follow the above, the distinction you draw doesn't go beyond liking something (a large group of headphones), and liking something more (a specific group of headphones).

I hope you see the point I'm making.

 

[Zekbone]

I'm still skeptical of all this precision. You can definitely hear the differences between the kind of peaks these kinds of biological differences create, and an EQ that corrects it, but the differences are still usually subtle. Even in my own sighted EQ vs EQ comparisons with naturally bass deficient openbacks, treble corrections are pretty subtle as long as bass is also corrected for both. And by treble I don't mean in the 6khz+ range, I mean 1khz. I struggle to even hear a difference between a 10khz shelf and randomly stabbing with a high Q filter at 12khz. And I would even say the difference between a 6db bass shelf (or 5.5db shelf as Oratory sets his presets to) and a 10db bass shelf is not a dealbreaker for the music I listen to, even though I use a 10db shelf because I can and want to. The only thing affected is the "slam" for lack of a better word. The tonality (timbre?) is still similar. The closest thing I can compare it to is changing the subwoofer volume setting on my parents 5.1 soundbar. I love turning it up a bit.

I don't say this to be anti EQ. I use EQ on every headphone I own, no matter where I am or going.

 

[Curvature]

I'm still skeptical of all this precision. You can definitely hear the differences between the kind of peaks these kinds of biological differences create, and an EQ that corrects it, but the differences are still usually subtle. Even in my own sighted EQ vs EQ comparisons with naturally bass deficient openbacks, treble corrections are pretty subtle as long as bass is also corrected for both. And by treble I don't mean in the 6khz+ range, I mean 1khz. I struggle to even hear a difference between a 10khz shelf and randomly stabbing with a high Q filter at 12khz. And I would even say the difference between a 6db bass shelf (or 5.5db shelf as Oratory sets his presets to) and a 10db bass shelf is not a dealbreaker for the music I listen to, even though I use a 10db shelf because I can and want to. The only thing affected is the "slam" for lack of a better word. The tonality (timbre?) is still similar. The closest thing I can compare it to is changing the subwoofer volume setting on my parents 5.1 soundbar. I love turning it up a bit.

I don't say this to be anti EQ. I use EQ on every headphone I own, no matter where I am or going.

High precision is not necessary in most cases, particularly not for stereo. I have found it useful to address idiosyncratic headphone resonances.

 

[listener650]

Biased of course both because Konstantin is a homie and we've been working on in-ear mic stuff together, but yeah this talk was both excellent as well as unusually brief (but in a good way) compared to most CanJam talks. Highly recommend anyone curious about headphone measurements check out the slides!

 

[oratory1990]

Let me know if there's additional questions

 

[Curvature]

Let me know if there's additional questions

Do you have any comments on the discussion in this thread so far? Are there any areas where there is a lack of clarity or general misunderstanding?

Is there anything you were surprised by in doing the research?

 

[oratory1990]

Random thought: Lasers can be good for measuring vibration and so are sometimes also used as microphones. Measuring response at the eardrum with a mechanical transducer is tricky because sticking things into people's ears without injuring them is tricky, not to mention you have to position it really accurately at the same time. Is there any such thing as a laser microphone that can measure the motion of the eardrum directly? I feel like that would be definitive in this arena.

I like the idea!
One issue with this comes to mind:
the ear canal is not straight, it is bent. On many individuals there is no direct line-of-sight from the eardrum. See example of a real-life ear canal below. (the eardrum is to the bottom-left, the ear canal entrance is to the top-right) The position and angle of the first and second bend is not identical for every person, and in many cases will be shaped such that a straight line from the eardrum to the canal entrance is not possible. The direct consequence is that we can not always look at the eardrum from the outside and hence also not point a laser to it. Meaning we would need to insert the laser emitter into the ear canal, obstructing it. An obstructed ear canal obviously affects the acoustics

We do have probe microphones that can be inserted into the ear canal, but those have a limited frequency range - because we can not insert them directly next to the eardrum. Audiologists regularly use those when fitting hearing aids, but they only insert them about 3mm away from the eardrum, limiting the accuracy to about 8 kHz (which is all that audiologists are interested in anyway, that's all that is needed for hearing hearing aids).

As a preference target, harman was found to not even rate that high and required custom bass and treble adjustments via their own research.

What do I even say other than: Please read the actual research results. This is not what happened.

3. Use frequency sweeps - correct obvious peaks/dips you actually hear.

I don't generally recommend using frequency sweeps to listen for peaks and dips - there is no requirement for a sweep to sound flat to your ears.
Listening to a sweep can tell you the frequency (and with some exercise, also the q-factor) of a resonance, but it does decidedly not tell you whether or not that peak should be reduced / whether that dip should be filled in.
Case in point: imagine an absolutely flat loudspeaker in a perfectly anechoic room - listening to a sine sweep on this loudspeaker will still create resonances in your ear depending on which angle you turn your head. We can measure those resonances! That's what the HRTF is. In the absence of context, you may or may not be able to actually hear those peaks when listening to a sine sweep - because the resonances are real (your ear shape creates them) even though the loudspeaker itself is completely flat.
You can find the frequency of peaks with a sweep, yes, but then use broadband signals (i.e. music) to find out whether those peaks need to be removed or not.
Peaks can be audible in a sine sweep and are expected, that in itself is not something that needs to be corrected.

I struggle to even hear a difference between a 10khz shelf and randomly stabbing with a high Q filter at 12khz.

That's actually not all that surprising! The rule of thumb is that in the top octave (above 10 kHz) the exact frequency response isn't nearly as important as it is in other frequency ranges, and it's mostly the total energy in that band that matters.
Which lines up exactly with your own experience.

Is there anything you were surprised by in doing the research?

I was actually surprised by how much the sound pressure on the B&K 5128 (The type 4.3 ear simulator) deviated from the sound pressure measured on humans.
It's marketed as a more accurate test fixture, but these tests don't support that.
For all over-ear headphones that I've tested, the Type 3.3 ear simulator gave more accurate results (read: gave results more in line with what we measured on actual humans) in the midrange and at high frequencies.
What was less surprising is that At low frequencies both ear simulators deviated from what we measured on humans in that they over-estimated the amount of bass. This is easily explained by the headphones achieving a better seal on the regular surface of the ear simulator's "cheek". This is why headphone manufacturers use leakage test setups (adding a controlled amount of leakage to the test fixture).
Something like this for in-ear headphones (the color-coded plugs can be removed and will each introduce a known impedance to the leakage path):

 

[FTB]

I like the idea!
One issue with this comes to mind:
the ear canal is not straight, it is bent. On many individuals there is no direct line-of-sight from the eardrum. See example of a real-life ear canal below. (the eardrum is to the bottom-left, the ear canal entrance is to the top-right) The position and angle of the first and second bend is not identical for every person, and in many cases will be shaped such that a straight line from the eardrum to the canal entrance is not possible. The direct consequence is that we can not always look at the eardrum from the outside and hence also not point a laser to it. Meaning we would need to insert the laser emitter into the ear canal, obstructing it. An obstructed ear canal obviously affects the acoustics
View attachment 557203

We do have probe microphones that can be inserted into the ear canal, but those have a limited frequency range - because we can not insert them directly next to the eardrum. Audiologists regularly use those when fitting hearing aids, but they only insert them about 3mm away from the eardrum, limiting the accuracy to about 8 kHz (which is all that audiologists are interested in anyway, that's all that is needed for hearing hearing aids).


What do I even say other than: Please read the actual research results. This is not what happened.


I don't generally recommend using frequency sweeps to listen for peaks and dips - there is no requirement for a sweep to sound flat to your ears.
Listening to a sweep can tell you the frequency (and with some exercise, also the q-factor) of a resonance, but it does decidedly not tell you whether or not that peak should be reduced / whether that dip should be filled in.
Case in point: imagine an absolutely flat loudspeaker in a perfectly anechoic room - listening to a sine sweep on this loudspeaker will still create resonances in your ear depending on which angle you turn your head. We can measure those resonances! That's what the HRTF is. In the absence of context, you may or may not be able to actually hear those peaks when listening to a sine sweep - because the resonances are real (your ear shape creates them) even though the loudspeaker itself is completely flat.
You can find the frequency of peaks with a sweep, yes, but then use broadband signals (i.e. music) to find out whether those peaks need to be removed or not.
Peaks can be audible in a sine sweep and are expected, that in itself is not something that needs to be corrected.


That's actually not all that surprising! The rule of thumb is that in the top octave (above 10 kHz) the exact frequency response isn't nearly as important as it is in other frequency ranges, and it's mostly the total energy in that band that matters.
Which lines up exactly with your own experience.


I was actually surprised by how much the sound pressure on the B&K 5128 (The type 4.3 ear simulator) deviated from the sound pressure measured on humans.
It's marketed as a more accurate test fixture, but these tests don't support that.
For all over-ear headphones that I've tested, the Type 3.3 ear simulator gave more accurate results (read: gave results more in line with what we measured on actual humans) in the midrange and at high frequencies.
What was less surprising is that At low frequencies both ear simulators deviated from what we measured on humans in that they over-estimated the amount of bass. This is easily explained by the headphones achieving a better seal on the regular surface of the ear simulator's "cheek". This is why headphone manufacturers use leakage test setups (adding a controlled amount of leakage to the test fixture).
Something like this for in-ear headphones (the color-coded plugs can be removed and will each introduce a known impedance to the leakage path):
View attachment 557205

Thanks for sharing your knowledge with us.