CHANGSTAR: Audiophile Headphone Reviews and Early 90s Style BBS
Lobby => Soapbox => Topic started by: funkmeister on August 28, 2013, 04:39:59 PM
-
I'm calling it! HRTF is BS for headphones. poo
My main operative theory here is that the sound needs to enter the ear as similarly to natural sound as possible. I don't want to take that too far with things like crossfeed, just from an FR standpoint. It's the point of entry that matters. The whole idea of trying to replicate a human ear canal and then measuring the resultant distorted signal inside of there, then correcting the distorted FR with a compensation curve is a load of BS and a waste of time, money, and other resources. At most we only need to compensate for boundary effects a bit in developing an ideal headphone curve.
Now... HRTF matters for IEMs because if you have been hearing with your own set of ears for your entire life up to this point, and I assume you have, then an IEM which is close to your eardrum needs to be tweaked to reproduce what your head and ear canal does to sound. That's why your individual HRTF matters for IEMs but a generic HRTF may not be good enough.
So, we just need to figure out what subtle rise and fall on the FR chart is needed to keep sound from a headphone in line with natural sound. I never really claimed that making things sound like good speakers was the way to go, but having heard what the Harman researchers have to say about it (thanks Tyll for sharing), I've adopted the thinking that it's the perfect place to operate in terms of research.
I made the case for rethinking HRTF in a few places about a year ago and people got defensive and made it into a fight because HRTF was untouchable and they went on trying to educate me all about it. Well... I think the whole concept needs to be rethought.
-
HRTF matters if headphones are measured with the microphone inside the ear near the eardrums. You are partly right though. If you think about it, we can get a more simple approximation bu putting the mic and measuring from outside the ear a cm or so. hmm lets wait for Purrin and the other measurement experts to chime in
-
It depends upon the measurement method on whether to apply "HRTF" or not. Most folks over at HF, when talk of HRTF was all the rage, seem to have confused HRTF with someone much simplier: FR compensation. I guess HRTF in its most simplistic form is a base compensation. An ideal behind HRTF is that the transfer function (I bet 99% of HF has no idea what a transfer function is in mathematical terms) should be three dimensional. One way in implementing this is to use individual impulse response transfer functions based on specific angular measurements in a spherical coordinate system. In practice, one would take limited set of measurements of different azimuth and altitude, and interpolate. The idea behind this is of course to to simulate a room / venue or provide directional cues with limited number of speakers. That's key. How HRTF got misapplied to the concept of very basic FR compensation, I do not know. Probably because it sounds fancy, and if one uses fancy words or acronyms, then one must be smart.
IEMs require a compensation to boost the a narrow area around 2-3kHz. When I spoke with UE, it was their opinion that the pinna boosted that region, so compensation was required for IEMs, which of course bypass the pinna. I do feel the same way in a subjective sense. (Note that the IEM measurements on this site already compensate for this - so the measurements you see are "perceptual" rather than "raw" measurements.) The raw measurements of the UERM do show a boost in this area. The ER4 also have a boost, although the I think it's overdone. IEMs which don't have that boost in the raw measurements seem to sound too laid back with vocals, particularly female vocals. This is not necessary a bad thing. The "BBC curve" used for speaker voicing pushed 3k back to lessen exaggerated room effects from the combination of the recording venue and the room used for playback. I find the BBC curve useful for live concerts, but not so helpful with studio recordings.
Speaker measurements are done with a microphone in free air. The v1 headphones measurements on this site where done in a simulated free-air environment using a porous sponge. These measurements were combobulated with another sealed plate measurement to arrive at the correct bass.
As for compensation used here, it's rather simple. I compare to references I consider neutral: my speakers and my audiophile desktop setup (both surgically EQ'd via digital PEQ to a target FR with a slight downward slope. If no one wants to believe me that this curve yields the "best" most neutral results, just ask the Harmon researcher Dr. Sean Olive, PhD.) I think I've posted a few listening position FR graphs of those setups. People can argue all they want about HRTF, compensation curves, diffuse field, AES papers, B&K recommended curves, etc. but at least I've provided a reference - and compensated as best as I could the measurements on this site to those references. Again, the measurements here are perceptual FR graphs. And finally, several folks here have actually heard my reference setups.
But yeah, in the context of how most folks at HF use the term HRTF with measurements, it's BS.
-
Transfer Functions and Fourier Transforms are closely related in general. However, some headphone Transfer Functions (and respective Fourier Transform) may be somewhat independent of HRTFs (specially with headphones that do not bypass the external ear).
It makes sense that realistic HRTFs are 3-D functions which may be unique to a particular individual. However, I believe performance of a sound reproduction system should be characterized relative to real world sound sources, which are independent of HRTFs. Note that speaker characterization is independent of surroundings (anechoic chamber) and head (with small mic a certain distance away from source).
One difference between speakers/sound-sources and non-IEM headphones may be that cans bypass the front of the head and fire directly to the ears. So I also believe some slight compensation (not the full HRTF measured at the ear canal) should be applied in the characterization of most headphones that do not bypass the ear. This is because many recordings target speaker reproduction. Acoustic impedance and anechoic chamber/plate emulation might be needed though.
IEMs bypass quite a few ear parts which are individual dependent, so some further compensation might be needed, and this compensation might not be 100% accurate for every individual... an average might be close. The ear might do some octave averaging to some extend, but dunno.
-
Actual HRTF is a 3D complex domain function, but in terms of headphone and IEM listening, only the 2D component matters, and usually only the one for 60 degrees to the side or front one, depending on which one is chosen.
The rest matters for spatializers or head tracking.
It might be that Fourier model of the head function is insufficient as the function is actually nonlinear.
Head itself is a minor part of the HRTF - the most important part is the outer ear.
Mono equalization by ear derives the frontal HRTF convolved with headphone/IEM response. Usually this is very close to the compensated measurement, perhaps with some resonant shifts and minor differences in upper midrange gain (around 3kHz) as well as ear canal resonance amount (around 6kHz).
There are still a few variables in headphones - distance from eardrum, central or off-center positioning of the driver which changes outer ear gain.
For IEMs, there's insertion depth and exact outer ear gain.
Both can also suffer from bad acoustic seal, like this:
(http://mountaininterval.org/photos/images/31-roll/29-cierva-cove-leopard-seal.jpg)
-
Actual HRTF is a 3D complex domain function, but in terms of headphone and IEM listening, only the 2D component matters, and usually only the one for 60 degrees to the side or front one, depending on which one is chosen.
The rest matters for spatializers or head tracking.
The problem I see is positional dependence though. It may not be a big problem, but sound may be affected by IEM insertion deepness and angle. I think this is a 3-D problem.
It might be that Fourier model of the head function is insufficient as the function is actually nonlinear.
The Fourier Transform itself is actually a linear operator. While the kernel of the Fourier Transform is a complex sinusoid deal, the transform itself is a linear projection of an arbitrary (and possibly non-linear time domain) signal into a set of orthonormal complex sinusoids.
An analogy could be done with the addition operator... One may add two non-linear functions, but the addition operator remains linear.
Mono equalization by ear derives the frontal one. It is usually very close to the average one.
There are still a few variables in headphones - distance from eardrum, central or off-center positioning of the driver which changes outer ear gain.
For IEMs, there's insertion depth and exact pinnae and outer ear gain.
Both can also suffer from bad acoustic seal, like this:
(http://mountaininterval.org/photos/images/31-roll/29-cierva-cove-leopard-seal.jpg)
:)p13 Yup, to me IEM characterization seems a little more involved.
-
In the overall scheme of things, I just like to keep things simple rather than worry about HRTF and other nonsense. In other words, how do the HD800s or <insert headphone here> sound different in relation to the HD650 or <insert other headphone here>, and do the measurements jive?
Those searching for the holy grail of absolute measurement perfection citing a myriad of authorities, methods, etc. are doomed to fail, or at the very least are not seeing the forest for the trees. It's like those fuckers looking for the "God" particle. What a fucking waste of human thought, time and money. What would be funny if after they "discovered" the God particle, some other random fucker physicist hypothesizes about a "Super-God" particle. And you know that's going to happen.
-
It's hard to really get perfection because even with speakers where the measurements are pretty much established, manufacturers still have different interpretations of how things should sound.
-
Exactly. While speakers can measure neutral according to the Stereophile method of using anechoic measurements - hanging the speaker up in the air and using a measurement mic 1 meter away - there's no guarantee they will sound even neutralish when in placed in most real rooms. This thing's anechoic measurement http://www.stereophile.com/content/sonus-faber-stradivari-homage-loudspeaker-measurements (http://www.stereophile.com/content/sonus-faber-stradivari-homage-loudspeaker-measurements) a was fucking god-awful bass monster in the small hotel room at THE SHOW. Anax I and quickly ran from the room in horror. But I bet in an auditorium sized room placed far far away from any walls, it would sound decent (still too much bass for my tastes).
Many studios pros and even Dr. Sean Olive (see Tyll's IF article) have described a modified curve (this modified speaker-in-room curve is essentially a tilted straight line with the bass end about 10dB above the treble end) at the listening position as being most preferred. It should be noted that I only brought in Sean Olive because some people worship him and require an authority. Literature on studio monitor EQ has long suggested this curve before Tyll's article mentioning Dr. Olive's research.
Even then, I personally use about a 7db curve (final curve is tuned by ear) with my references, possibly because my living room and work room are more reverberant than most studio environments. At the end of the day, at least a few speaker manufacturers do know what they are doing. There were four / five setups at THE SHOW 2013 which I felt were within reasonable bounds of neutrality / listenability and I didn't really hear anything which sounded nasty like an Ultrasone or HD700 except for this wide baffle VoaAtiv speaker which was a tonal balance disaster:
(http://www.instablogsimages.com/images/2011/01/05/voxativ-ampeggio-due_YzeFS_48.jpg)
-
Yup. Remember quite a few Ultra-Uber-Godly-Reference speakers at THE SHOW had be equalized, moved around, and even blessed with the steel balls of Hephaestus due to room and driver limitations.
In the end I think one just has to work with what is available in the real world and get the most of it.
Relative impressions and measurements are very useful IMO. In all cases I think one needs to have real references (actually hearing the phones/speakers) to base ones own preferences. The proverb goes "There is not substitute for experience".
-
Exactly. While speakers can measure neutral according to the Stereophile method of using anechoic measurements - hanging the speaker up in the air and using a measurement mic 1 meter away - there's no guarantee they will sound even neutralish when in placed in most real rooms.
Add to that... calibrated speakers with different dispersion characteristics (ex. magnepans vs typical box speakers) measured at 1m ala stereophile will never generate the same raw curve in the ear with each other unless maybe the measurement was done at the same distance of 1 meter. If an in-room response at 2 to 3 meters were made with those speakers with mics in the eardrums, they would surely give us a different resulting raw curve.
By the way, about speaker measurement, it is actually also even a mystery how they came up with the frequency response of the magneplanar speaker for example since with the stereophile 1m method, they dont measure nearly as flat vs typical box speakers yet they usually sound more natural in most environments I heard them in. I wonder if they tuned the maggies by ear or by measuring from the sweet spot of a typical room.
-
Exactly. While speakers can measure neutral according to the Stereophile method of using anechoic measurements - hanging the speaker up in the air and using a measurement mic 1 meter away - there's no guarantee they will sound even neutralish when in placed in most real rooms.
Add to that... calibrated speakers with different dispersion characteristics (ex. magnepans vs typical box speakers) measured at 1m ala stereophile will never generate the same raw curve in the ear with each other unless maybe the measurement was done at the same distance of 1 meter. If an in-room response at 2 to 3 meters were made with those speakers with mics in the eardrums, they would surely give us a different resulting raw curve.
By the way, about speaker measurement, it is actually also even a mystery how they came up with the frequency response of the magneplanar speaker for example since with the stereophile 1m method, they dont measure nearly as flat vs typical box speakers yet they usually sound more natural in most environments I heard them in. I wonder if they tuned the maggies by ear or by measuring from the sweet spot of a typical room.
Some interesting thoughts from Linkwitz and Mike Gough on the matter. Look at what JA measured on the previous page for context.
http://www.stereophile.com/content/magnepan-magneplanar-mg36r-loudspeaker-more-comments
-
Since low tones reflect more and higher tones are more easily absorbed, how come we don't have colorized charts to display a headphones characteristics in that regard? Well... CSD does that to a certain extent but I'm thinking in terms of 2D and with and indicator as to where the crossover point should be. Take that and compare to the CSD and you'd have a way figured out to immediately discern and improve on structural deficiencies.
-
Those searching for the holy grail of absolute measurement perfection citing a myriad of authorities, methods, etc. are doomed to fail, or at the very least are not seeing the forest for the trees. It's like those fuckers looking for the "God" particle. What a fucking waste of human thought, time and money. What would be funny if after they "discovered" the God particle, some other random fucker physicist hypothesizes about a "Super-God" particle. And you know that's going to happen.
I disagree strongly, but that's quite off-topic. It's the most important discovery of the 21th century so far. Btw, most physicists hate the nickname "God particle" and Lederman wanted to call it the Goddamn particle. It has nothing to do with fairy tales.
Since low tones reflect more and higher tones are more easily absorbed
Well, isn't it kinda the other way around with headphones? At low frequencies it works more like a hydraulic brake while at high frequencies there are waves that are being reflected from the ear cup, pinna, ...
What do you mean with crossover point?
-
Those searching for the holy grail of absolute measurement perfection citing a myriad of authorities, methods, etc. are doomed to fail, or at the very least are not seeing the forest for the trees. It's like those fuckers looking for the "God" particle. What a fucking waste of human thought, time and money. What would be funny if after they "discovered" the God particle, some other random fucker physicist hypothesizes about a "Super-God" particle. And you know that's going to happen.
I disagree strongly, but that's quite off-topic. It's the most important discovery of the 21th century so far. Btw, most physicists hate the nickname "God particle" and Lederman wanted to call it the Goddamn particle. It has nothing to do with fairy tales.
For amplifier and DAC characterization (on their own), setting flat frequency response and practically non-existent levels of distortion for a target set of loads and dynamic ranges maybe yield an arbitrary absolute one dimensional performance goal (maybe two for stereo). It is still a little arbitrary since it fails to account for whatever is down the chain (headphone/speaker, room/head...), but it maybe generic enough.
Transducers however seem to interact with the medium quite a bit, the problem is basically three dimensional and not completely individual/room and headphone/speaker type independent. There are different design goals: radiation pattern, coherence, distortion, sensitivity... Relative impressions and measurements become useful in such cases, while absolute may help set a reference.
In speakers it maybe possible to find a particular build that would perform well +/- 45 degrees horizontal and vertical in anechoic conditions, but perform poorly on a particular (and not necessarily small) set of real world rooms... One could equalize to perfection at one location and fail miserably in all the other non-sweet-spot locations... One may not be able to equalize to perfection anywhere in some rooms and with some headphones as well...
It doesn't mean we shouldn't try to figure things out, but resources should be wisely allocated... Perhaps instead of figuring out absolute perfection on every single room/headphone geometry condition and location, some research could be done on finding an independent of head/room (a la anechoic room) and convenient standard metric for headphone characterization, along with it's subjective sound quality perception variance, which seems to be getting current research attention attention anyway. Perhaps some research on particular frequency range masking would be useful... dunno.
Since low tones reflect more and higher tones are more easily absorbed
Well, isn't it kinda the other way around with headphones? At low frequencies it works more like a hydraulic brake while at high frequencies there are waves that are being reflected from the ear cup, pinna, ...
What do you mean with crossover point?
Probably the ones used for those Magnepan transducers from the Stereophile linky.
-
I disagree strongly, but that's quite off-topic. It's the most important discovery of the 21th century so far. Btw, most physicists hate the nickname "God particle" and Lederman wanted to call it the Goddamn particle. It has nothing to do with fairy tales.
We'll see. The discovery of Higgs and validation of the standard model, unification of electromagnetic / weak forces, how particles get mass, etc. - it better result in warp drive or a photon torpedo we can ram up Assad's ass - otherwise's it's just intellectual masturbation for theoretical physicists. I suspect there will be no practical application of this scientific discovery because of the high energies involved at which these models work, unlike quantum mechanics, e.g. tunneling, wave functions, etc.
-
OMG, and how gravity tightly fits inside all of it...(http://www.clker.com/cliparts/P/r/K/Y/b/5/white-splash.svg)
-
I've long been preaching my theory among those who will take time to hear my theories that the 4 forces in the universe are all the same force which simply abides behavior appropriately.
Anyway, the more I think about my proposed graphing techniques the more arbitrary my crossover point seems. I can't quite work out a model with only two main frequency groups.