Equal loudness curve testing

[Quaintative]

Quote:

Do you mean magnitude and phase response of the frequency response and distortion? Distortion definitely can be a problem when testing the equal loudness of very low frequency tones, which have to be played at very high SPL.

Well, what I was thinking was how frequency curves are also dependent on volume. You could take the ISO loudness curve and stack them together to build a representative loudness "plane". I think you could do the same thing with a given headphone.
 
I'm not so sure if that's what you mean by magnitude and phase response/frequency response and distortion. I suspect the latter.

 

[jaddie]

Quote:

Well, what I was thinking was how frequency curves are also dependent on volume. You could take the ISO loudness curve and stack them together to build a representative loudness "plane". I think you could do the same thing with a given headphone.

Yes, that's the idea.  The one non-trival issue is, determining the exact SPL at the hear moment by moment, spectrally (not over-all) and making the appropriate correction based on that.  Sorry to keep beating the horse here, Audyssesy "amp" does exactly this with Dynamic EQ, and for specific headphones.  The app is a buck. Or try it free with the Songza app.

 

[xnor]

Quote:

Well, what I was thinking was how frequency curves are also dependent on volume.

That's why I wrote you have to choose a curve/SPL.
 
 
It all depends on what you want to achieve. If you want to flatten the perceived frequency response of your headphones try the process I've described above.
On the other hand, if you listen at low SPLs and your headphone sounds mids-heavy then compare the sound sig at higher SPL to the one at low SPL and equalize accordingly.
Lastly, see what jaddie wrote if you want a more elaborate/dynamic/flexible loudness effect.

 

[jaddie]

Quote:

That's why I wrote you have to choose a curve/SPL.
 
 
It all depends on what you want to achieve. If you want to flatten the perceived frequency response of your headphones try the process I've described above.
On the other hand, if you listen at low SPLs and your headphone sounds mids-heavy then compare the sound sig at higher SPL to the one at low SPL and equalize accordingly.
Lastly, see what jaddie wrote if you want a more elaborate/dynamic/flexible loudness effect.

It's a small-ish point, and not disagreeing at all, but choosing a curve/SPL would result in an error.  The curve only represents hearing response at a given SPL vs frequency.  For example, if you looked at an 85dB SPL curve vs a 50dB SPL curve, the correction needed would be the 50dB curve minus the 85dB curve, assuming 85dB was the mix reference level, and all material in the spectrum at any moment was presented at the exact same SPL...which, of course, it wouldn't be because it's changing all the time.
 
No single curve represents the specific correction at all, because in music or any material the exact correction is the difference in response between the mix level and the reproduced level at a given frequency, and actually using a curve would be wrong, unless all stimulus in the spectrum was presented at exactly the same level.  This is also why the correction cannot be fixed, and must be dynamic, and also why doing this at all had to wait for DSP to catch up.  There's simply no good way to realize this in the analog world.  

 

[Quaintative]

Not to mention the HTRF distorting things a little bit more....

 

[jaddie]

Quote:

Not to mention the HTRF distorting things a little bit more....

HRTF relates only to angle of incidence, and doesn't change dynamically with SPL. It can be considered fixed for a given angle, over a range of variance for individuals physical characteristics. 

 

[xnor]

@jaddie:
In the first case, the chosen curve is only used with pure tones that of course should be reproduced at precisely the right SPL. Here the equal loudness contour is just an intermediate tool to try to equalize FR balance problems in the headphone.
 
In the second case, I wouldn't just subtract one curve from the other but compare the sound signature using music at the two SPLs.
 
In both cases some error is going to be unavoidable.
 
 
I'm not a fan of dynamic EQ filters during reproduction, but I have not used the app you mentioned. Is the dynamic part only an adjustment to the volume control's position, or does it incorporate the signal level also?
During the quiet passages of a concert nobody's gonna turn up the bass either.

 

[jaddie]

Quote:

@jaddie:
In the first case, the chosen curve is only used with pure tones that of course should be reproduced at precisely the right SPL. Here the equal loudness contour is just an intermediate tool to try to equalize FR balance problems in the headphone.
 
In the second case, I wouldn't just subtract one curve from the other but compare the sound signature using music at the two SPLs.

Subtracting one from the other results in the differential correction curve, but ONLY at one SPL for each frequency group.  Still doesn't work, but I mentioned it because so many people think the equal loudness contours are the correction, or inverse correction curves themselves. They are not.  They are the curves that show the difference in SPL to provide equal loudness of tones or band limited noise.
Quote:

I'm not a fan of dynamic EQ filters during reproduction, but I have not used the app you mentioned. Is the dynamic part only an adjustment to the volume control's position, or does it incorporate the signal level also?
During the quiet passages of a concert nobody's gonna turn up the bass either.

Audyssey's "Dynamic EQ" is not just a filter changed dynamically.  It's far more complex, and the action arrived at through extensive experimental research that generated many terrabytes of data using real human listeners.   You won't find much about how they did it, but the general description of what they do is here:
 
http://www.audyssey.com/audio-technology/dynamic-eq
 
If you want more about how the research was done, PM me. 
 
When the dynamic filtering action is based on real human hearing response, as a correction for the spectral sensitivity differences cause by offset play level, it actually works.  It's not an over-all gain change at all, it's not compression either.  It's selective response correction applied dynamically, and this is the key: with full knowledge of the actual SPL at the ear.  When done right, the control action is transparent, just sounds "right".   Audyssey Dynamic Volume is something entirely different, also well done, but not with the same goal, and not found in the "amp" app.
 
Concerts need no correction, they are the "reference".  If you played a concert recording monitored and mixed at concert levels in a calibrated control room, and played on your calibrated system at original concert levels, no correction is required.  Loudness correction is only needed when play levels are different from mix levels.

 

[xnor]

Quote:

When the dynamic filtering action is based on real human hearing response, as a correction for the spectral sensitivity differences cause by offset play level, it actually works.  It's not an over-all gain change at all, it's not compression either.  It's selective response correction applied dynamically, and this is the key: with full knowledge of the actual SPL at the ear.  When done right, the control action is transparent, just sounds "right".   Audyssey Dynamic Volume is something entirely different, also well done, but not with the same goal, and not found in the "amp" app.
 
Concerts need no correction, they are the "reference".  If you played a concert recording monitored and mixed at concert levels in a calibrated control room, and played on your calibrated system at original concert levels, no correction is required.  Loudness correction is only needed when play levels are different from mix levels.

I'm still not convinced because for example if you play the recording at a 20 dB lower level you want a 20 dB difference loudness correction curve (as described above) regardless of the recording's level. For example, when there are silent (e.g. 30 dB softer) passages you don't want a 50 dB difference loudness correction curve.

 

[jaddie]

Quote:

I'm still not convinced because for example if you play the recording at a 20 dB lower level you want a 20 dB difference loudness correction curve (as described above) regardless of the recording's level. For example, when there are silent (e.g. 30 dB softer) passages you don't want a 50 dB difference loudness correction curve.

Playing a recording 20dB lower doesn't result in a 20dB difference in the correction, even at the extremes.  Studying the curves, if you assume a reference at 85dB SPL mid-band, then drop 20dB, the correction at 20Hz is about 10dB, and progressively less as you move to the mid band.  The correct never comes close to 50dB at any practical listening volume level, in fact, probably maxes at 20dB at 20Hz at 40dB SPL mid-band re: mix reference.  Now, 20dB is not small, but considering that play level is skewed 45dB from mix level, and averaging at 40dB SPL, and the correction lets you hear the entire mix spectrally balanced, it's a significant advantage.  
 
But no amount of discussion of the principles will ever beat just listening.  If you have an iOS device, try the Songza app for free, or spend a buck on "amp".  All in all, very little lost if things don't go well.  But if you do like what you hear, you've found a great app, and may actually want to try the same processing in your AVR some day.  The audition would be to listen at a reference level, like around 80dB SPL or so, then with Dynamic EQ on, drop the play level 20dB or so and listen for spectral balance.  Then, without touching the volume, turn the processing off.  You'll get the idea pretty quickly.
 
Personally, the only reason I don't use it all the time is my favorite player is a 160gig iPod with my entire library on it, and "amp" won't run on an iPod.  So to use "amp" I have to load up a selection of tracks to my iPhone or iPad, and listen to that little library with "amp".  
 
I'm not a big fan of Songza, I guess I'm too old. 

 

[xnor]

Quote:

Playing a recording 20dB lower doesn't result in a 20dB difference in the correction, even at the extremes.

I wasn't saying that. With '20 dB difference loudness correction curve' I mean, for example, the result of the 85 minus 65 dB curve. What precise amount of boost that causes at 20 Hz doesn't really matter to what I wrote.
 
Let me try to explain again.. At the concert a 1 kHz sine wave is played at 85 dB SPL. At your home the volume control is set so that this tone is reproduced at 65 dB SPL. Let's assume there's also a 20 Hz tone at the same SPL and that the 85-65=20 dB 'difference loudness correction curve' results in +10 dB at 20 Hz, so this tone is actually reproduced at 65+10=75 dB SPL.
Now at the concert if the level of the 100 Hz dropped by 10 dB to 75 dB SPL it should also drop to 55+10=65 dB SPL at your home and not some higher value due the dynamic filters detecting a lower signal level and using a 30 dB 'difference loudness correction curve' instead.
 
The 10 dB difference should be preserved at reproduction.
 
In other words, I see no reason why the filters shouldn't be static except for the overall track or album level and volume control position (and speaker sensitivity ...).

 

[jaddie]

Quote:

I wasn't saying that. With '20 dB difference loudness correction curve' I mean, for example, the result of the 85 minus 65 dB curve. What precise amount of boost that causes at 20 Hz doesn't really matter to what I wrote.
 
Let me try to explain again.. At the concert a 1 kHz sine wave is played at 85 dB SPL. At your home the volume control is set so that this tone is reproduced at 65 dB SPL. Let's assume there's also a 20 Hz tone at the same SPL and that the 85-65=20 dB 'difference loudness correction curve' results in +10 dB at 20 Hz, so this tone is actually reproduced at 65+10=75 dB SPL.
Now at the concert if the level of the 100 Hz dropped by 10 dB to 75 dB SPL it should also drop to 55+10=65 dB SPL at your home and not some higher value due the dynamic filters detecting a lower signal level and using a 30 dB 'difference loudness correction curve' instead.
 
The 10 dB difference should be preserved at reproduction.
 
In other words, I see no reason why the filters shouldn't be static except for the overall track or album level and volume control position (and speaker sensitivity ...).

Hearing sensitivity an low frequencies is non-linear with respect to SPL. Play volume offset puts you on a different virtual curve for the volume of every note.  If play levels and concert levels are matched, during playback hearing non-linear response tracks the original.  If play volume is offset, then you are not only on a different virtual curve, but a different point in the non-linear response curve of hearing at low frequencies.  The difference between a static curve and one that is influenced by specific SPL at specific frequencies is not large, but is significant. It's part of what makes the system transparent vs artificially bass-heavy or bass-light.  The system is not trying to sevro-out the normal perception of reduced bass as concert performance volumes change, it's just compensating for play volume offset putting you on a different part of the hearing non-linearity curve.

 

[xnor]

I see what you mean, but subtracting the 85 from the 65 phon curve and the 65 from the 45 phon curve and comparing the two shows that the differences are pretty similar (max ~0.3 dB difference at 100 Hz). This error gets about 1.5 dB (again only at 100 Hz) if you go down another 20 dB vs. the 65-85 difference.
 
This is of course using the ISO 226:2003 curves. Audyssey may be using different curves which are a lot more non-linear, but I doubt it.
 
But I can see now how a subtle dynamic effect could help.

 

[jaddie]

Quote:

I see what you mean, but subtracting the 85 from the 65 phon curve and the 65 from the 45 phon curve and comparing the two shows that the differences are pretty similar (max ~0.3 dB difference at 100 Hz). This error gets about 1.5 dB (again only at 100 Hz) if you go down another 20 dB vs. the 65-85 difference.
 
This is of course using the ISO 226:2003 curves. Audyssey may be using different curves which are a lot more non-linear, but I doubt it.
 
But I can see now how a subtle dynamic effect could help.

Audyssey conducted their own research, didn't base the algorithm on anything pre-existing.  Without giving away the farm, the research involved a large number of human subjects, and they got more data than anyone ever had before.  

 

[Quaintative]

Time to find a research lab with highly accurate microphones, a proclivity for high-end headphones and an fMRI......