Sound Science Questions

.

 

This is an interesting area. I think we know some of these things. This will be out of order, but I'm on my phone:

Frequency response definitely affects timbre, and the partials are part of the reason. So it is true that an incorrect FR would misrepresent timbre.

Headphones sound bad if they measure flat (raw, uncompensated).

We know the reason for this is the geometry of the head and ear.

To compensate for the ear geometry people design compensation curves called head related transfer functions (HRTF). No one agrees on the perfect HRTF for headphones yet, but they seem to be getting closer.

The human brain is an amazing signal processor. Being able to identify different pitch levels, like being able to say "the bass is low", or "the mids are forward", is a transformation from time/amplitude domain to frequency/amplitude domain. We use FFTs to do this in electronics, but our brain can do it in such detail that we can pick out the different instruments that make up a sound. That's remarkable to me.

 

Mm... I just wanna say... don't bank too much on FR, or distortion, or etc...

1) I have measured two headphones (HD600 vs HD650) with the exact same FR and distortion figures. They don't sound the same at all. Timbre or otherwise.

2) Headphones don't sound bad when they measure with a "flat" FR. In fact, I'd wager that a headphone that sounds relatively flat will be more likely to pull out more details than one that doesn't. Some exceptions apply, but this is true for the norm.

3) Head and ears are important, but the housing of the headphone is important as well.

4) HRTF and other such "transfer function" will only work reasonably well for certain headphones due to problem #1

So... that's all just to say... trust in your heart (and ears).

 

I don't mean headphones that sound flat. I mean headphones that measure flat. The human ear expects gain in the upper mid range, and headphones that measure flat sound sucked out in the mids. Like most planars.

 

I think we should keep in mind that instruments aren't ideal objects capable of portray a mathematical set of tones and overtones at exact levels.
And we aren't tools waiting for specific set of tones and overtones at exact levels.

Pick two different violins and they'll sound different.
Pick a violin and give it to a violinist, then give it to other violinist and the sound will be slightly different as well.
Pick a violin, a violinist and a concert hall, then keep the violin and the violinist but change the concert hall and yep, different sound.
Pick a violin, a violinist and a concert hall, sit at row 1 and enjoy, then go back to row 20 ------> different sound

Still identifiable violin-sound in all cases.

It's esperable that violins sound like violins despite slight differences in frequency response among different headphones.
Well, we are even capable of identifying a violin when played through a phone call...

So violins (and instruments in general) have a very, very distinctive sonic pattern, that allows plenty of deviations without overlapping with the sound of other instruments. (Let very similar instruments aside)

FR changes will surely change the relative strength of partials but won't change the specific partials.
Then different headphones have different harmonic (+ other sorts of) distortion, so that might also play a role tweaking the sound a bit.

Last but not least, there are no perfect recordings so you'll find distortion (FR, HD, etc.) there as well.

You can check a few graphs here:
A Harmonic Spectrum Analysis of Various Instruments

 

The timbre of instruments is mostly contained in the initial attack or impulse. Studies have shown that if this part of the sound envelope is cut off, it becomes difficult to differentiate unique sound signatures. As others have pointed out, that sound signature is determined by the frequencies and relative amplitides of the harmonics over the fundamental.

 

If I recall, I read about the timbre-attack thing in This is Your Brain on Music by Daniel Levitin. Not positive, though.

 

Eh... I did say "measure flat".

But anyway, just to say... I think your frame of reference is a bit skewered.

Most planars in fact do not measure flat, and many measure abysmally sucked out in the mids and very V-shaped.

The only full-size headphones I am aware of on the market that measure on most systems relatively flat from 200Hz all the way up to about 10KHz are actually HD600 and HD650. All else are actually not as flat, no exception.

The Orpheus HE90 (and maybe the Orpheus 2 HE1?) are probably the other full-size that has a relatively flat frequency response up there.

Otherwise, you're not really experiencing "flatness" with planars.

 

I don't think it is so much that our ears have a lower resolution, but that we measure sounds differently from electronic gear, in that our brain does not usually measure sound against a fixed reference as do artificial measuring systems but rather against what is currently present, so unless a person goes through adequate training (and I make no definite claims on how that might work out) he or she is unlikely to identify sounds in terms of absolute values.

 

Found this old paper that might be of interest: https://drive.google.com/file/d/0By8mqQEExakOZGJwQW1TRHM0YVE/view?usp=sharing
Can't believe the publisher still charges $30 for it even though it is more than 80 years old.

Edit: Might be better if I let you guys access the folder instead since I might be adding more papers that I find.
https://drive.google.com/folderview?id=0By8mqQEExakOeThwNUdqQ3MyWmc&usp=sharing

 

Hmmm, I'm not sure why we are not seeing eye to eye on this - I suspect that we are somehow talking about two different things. So here is the hd650 measurement from Tyll's site (ignore the isolation graph on the right, I just wanted to include the name of the HP in the image):




It's important to note that the gray lines are the raw, uncompensated measurements. Definitely not flat.

Note the big bump between 2-5k. This bump is intentionally tuned like this by Senn, very similar to the Harmon curve. It is needed because when you put a speaker right up to your ear (a headphone), the natural gain caused by the shape of the ear gets largely bypassed, so the headphone is tuned to raise that area on your ear's behalf, simulating what it would have done in a normal sound field (not against the ear).

The red/blue lines above are the FR after having applied an HRTF - in this case this is Tyll's compensation curve.

Hope that helps.

 

Now here's a headphone that measures nearly flat (again look at the raw measurements, the gray lines):



See how that 2-5k bump that was present for the hd650 has been bulldozed over here? This headphone has sucked out mids, because the FR plot is flattish between 2-5k. Also, it will sound bright (or at least the treble will be sharp) with the treble peaks extending that way.

 

I think you're perspective is skewed by the fact that all the graphs posted on this site have a transfer function applied - now raw response. This is a source of annoyance for me, for exactly the reason you demonstrate. As has been stated elsewhere on the site in no uncertain terms, there is no one curve that is "correct" or "neutral".

 

Yeah, I wish people would simply not apply compensation curves.

 

I think it's better to have both, a lot of people can't interpret raw measurements and would think a headphone measuring 'flat' on a raw measurement is 'good'. I mean it would be good if our ears actually worked like that, but they don't.

 

Flat (raw) on a measurement rig that includes a Ear is not the same as Flat (raw) on a minimalistic measurement rig (no Ear simulator) like most SBAF's measurement rigs.

Both Raw and Compensated graphs (Ear or not) can be tricky in objective terms.
It's always wise to conceive graphs as comparative tools

 

In my view, compensation curves can be useful for comparative purposes.

My rig is different than most rigs here and thus it yields slightly different frequency response plots.
If I publish uncompensated measurements they will be useless unless you want to compare two of my measurements.

With that in mind, I worked out a compensation so my frequency response plots match Marv's plots (using the well known HD650).

Now it's possible to compare my measurements with most measurements here.

In strict objective terms, no one really knows how HD650 would measure on an ideal measurement rig (or a less than ideal measurment rig + the perfect compensation).
I don't think that's achievable because we all have different body, brain, and preferred listening levels.

So we are left with the possibility of comparative analysis.

 

Nice, good call!

 

Fair enough, but that doesn't make "HD6x0 measures flat" an accurate statement.

 

it is...

Because when you measure without any acoustical obstruction (ear shape?) and it's "flat", as in... there's about an equal amount of energy that the ear receives at any frequency...

Then it is indisputably flat.

Tyll's method of measurement is not wrong, but notice how "flatness" on his graph ONLY APPLIES to the ear shape that he uses, and nothing else. Since people have different ear shapes, it is impossible to deduce how a single headphone sounds like to any other person by just looking at Tyll's graphs. This is why he has a transfer function to compensate for the frequency response, and also why "flat" even with the transfer function would not necessarily sound good (this is explained by Tyll already... in one of his articles)

In contrast, a "flat" graph using the SBAF method will necessarily sound very close to a set of speakers that also measures flat at the same location. I know, because I have tried.

Also, notice how Tyll has consistently noted the HD600/650 as being tonally correct? That's not a coincidence. These headphones have been the reference for a long time for many people for a reason. But... of course, you are free to believe otherwise.

Edit: Just to add to this: I don't think Tyll's measurements are wrong (as mentioned above). But you gotta look at them the other way. His measurements are good when you want to look at other aspects, such as square waves and impulse response.

Otherwise, it's been discussed ad nauseam how Tyll's frequency response graphs look so darn weird. Tyll himself has said that he wanted to change the compensation and even the method, but because changing would necessarily invalidate all of his current data, he's stuck with his rig for a long time now. His graphs should be used more for comparative purposes between the different headphones he has measured, rather than as a tool to assess absolute tonal balance or anything of the sort. So talking about "flat" with his raw graphs is kinda... pointless.