[olsenn]

I have been looking at frequency response graphs for various headphones of different types (dynamic/orthodynamic/electrostatic), forms (closed/open/circumaural/supraaural), and price ranges, and it appears that not a single headphone ever made is perfectly linear (+-2db) from 1KHz to 10KHz... most don't even make it to 5KHz.
 
Why is it that headphones seem to be able to perform well up to 1K and not do any better than that? Even headphones like Grados which are known for having high mids and treble, are wavy and full of peaks/trophs? Does anyone know of a headphone that is linear from 1K - 10K... even if it lacks in the bass department?

 

[TMRaven]

It has to do with the compensation used.  It makes the measurements not the most reliable after about 2khz.  Tyll has his own interpretations of what truly neutral is for his graphs, and they vaguely resemble an HE-500.
 
Golden Ears and Purrin use different compensation (I'm not sure what Golden Ears does, but Purrin references off his neutral measuring speaker setup) so whatever is linear on his graphs is linear for the headphone.

 

[olsenn]

Quote:

It has to do with the compensation used.  It makes the measurements not the most reliable after about 2khz.  Tyll has his own interpretations of what truly neutral is for his graphs, and they vaguely resemble an HE-500.
 
Golden Ears and Purrin use different compensation (I'm not sure what Golden Ears does, but Purrin references off his neutral measuring speaker setup) so whatever is linear on his graphs is linear for the headphone.

 
I really don't understand this! I know that Tyll uses an expensive dummy head with precision microphones placed inside what are essentually human ears (whatever the microphone hears is EXACTLY what we hear!) If he connects the microphone output to his audio spectrum analyzer, whatever the graph shows is what you get.
 
By definition, "natural" sounding is linear! Unless the microphones aren't accurate above 2KHz, then natural sound would be a straight line from 20Hz to 20kHz. The HE-500 dips at 1kHz. It is the second closest to linear that I have seen (first prize goes to Orpheus).

 

[TMRaven]

No, it's more complicated than that.  The shape of the ear amplifies certain aspects of the frequency response, so you can't treat headphone measurements the same as say, speaker measurements.  There's also a role in how the brain compensates for this phenomenon as well.
 
In short, the majority of it falls on what compensation the measurement is using.

 

[olsenn]

Quote:

No, it's more complicated than that.  The shape of the ear amplifies certain aspects of the frequency response, so you can't treat headphone measurements the same as say, speaker measurements.  There's also a role in how the brain compensates for this phenomenon as well.
 
In short, the majority of it falls on what compensation the measurement is using.
 

 
I don't think you understand my point. The shape of the ear is accounted for in that Tyll's measurements are taken with the microphones in rubber earlobes that are shaped exactly as is the human ear. ALL aspects of the frequency response get affected by these rubber ears in the SAME way that they do on an actual human head. There is no need for compensation.
 
As for brain compensation, that is ridiculous. We do interpret sound with our brain, but if the sound produced by headphones is the SAME as the sound produced by the singer (as in air molecules are moving around the ear with the same force and frequency) then our brain will interpret that sound in the same way... our brain interprets sounds produced by headphones the same way as all other sound... the headphone just doesn't produce sound the way that they should 

 

[TMRaven]

If all this were true, there'd be no reason for us to use the compensated graphs provided by Tyll, we'd be using all of the raw graphs, as shown as gray frequency response below every blue and red one on his measurements. 

 

[olsenn]

Darn, now that confuses me. So how were the compensation algorithms determined, or is it a total guess based on listening experience?
 
I'm assuming Tyll's coloured graphs represent the compensated graph, and a perfectly linear one of these is ideal (for natural sound)? If Tyll says the HE-500 comes as close as any to produces "natural" sound, then I guess to rephrase my origonal question, does anybody know why no headphone is linear on Tyll's compensated graphs above ~1KHz? It seems like technology has a long way to go if the HE-500 is the best we have, haha.

 

[TMRaven]

In the article Tyll wrote explaining his FR measurements, seen here, he also links to an explanation of the HRTF he uses.
 
As I said before, Tyll himself doesn't think that perfectly linear on his compensated graphs is a perfectly linear response.  Time and time again he's stated that he thinks a slightly shelved upper midrange/lower treble in his compensated graphs is ideally more flat.

 

[AgentCOPP1]

Quote:

By definition, "natural" sounding is linear! 

That's not necessarily true. The human brain has the ability to dictate what sounds natural or not, and linear might actually sound very unnatural. Linear sometimes can sound very hollow and bland, so obviously it isn't always natural. Even if a headphone is truly linear, by the time the sound waves hit your ear drum, the signature has already been changed due to the physical properties of certain frequency ranges. That drop that you see around the mid-upper frequency range is simply because sound in that range does not resonate as well in the human ear, making the dummy head record a drop in that range.

 

[TMRaven]

No that's the exact opposite of the truth.  The ridges in the ear amplify the upper midrange part of the frequency.  Take a look at the raw graphs of Tyll's measurements.  That is what the microphone is hearing before compensation is applied.

 

[AgentCOPP1]

Quote:

No that's the exact opposite of the truth.  The ridges in the ear amplify the upper midrange part of the frequency.  Take a look at the raw graphs of Tyll's measurements.  That is what the microphone is hearing before compensation is applied.

I'm not talking about the upper mid. I'm talking about mid-upper, like around 10000. I don't know if mid-upper is the correct term, but that's what I'm talking about.

 

[TMRaven]

Quote:

I'm not talking about the upper mid. I'm talking about mid-upper

 
 
Yes, it could be considered the right term, just easily confused with upper-mids, like I did.
 
I havn't noticed too much of a recession in most raw graphs around 10k, but it's true that it doesn't quite amplify as strongly as the 3-5khz range.

 

[lostmage]

Quote:

I'm not talking about the upper mid. I'm talking about mid-upper, like around 10000. I don't know if mid-upper is the correct term, but that's what I'm talking about.

"Mid-treble" maybe?