[tinyman392]

The Effects of Small Changes over a Frequency Range
 
OK, up until now, it has been widely believed that a 1 dB change could not be heard, and the test to this pointing users to test it themselves by increasing the entire amplitude of a sound by 1 dB…  That data then goes on to FR graphs where we are widely made to believe that these same changes will make no audible change…  Well, my curiosity struck me like this…  Any instrument will provide more than one frequency of sound, some will product thousands of frequencies in a single note.  So why are we comparing the change in one frequency instead of change in multiple…  Because of the idea that a change in a single frequency isn’t large enough to be heard, that same change in multiple isn’t large enough either.  We make this assumption, we use this assumption.  Well, is it the right assumption, or are we really making an ass out of u and me? 
 
Please note that for all tests done, I used the EQu app available for iOS to get more accurate EQuing that allowed me to do more.  Although fully accuracy couldn’t be done in the first part of this test, the error created actually works against me here (allowing us to hear more change; you’ll see why it works against me in a bit).  However, let’s look at what this entire test will look at:

1. The effects of EQing a single frequency 4+ dB
2. The effects of EQuing a range of frequencies 1 dB up
3. The effects of EQuing a range of frequencies 1 dB up in some areas, down in others

 
Materials that I used for this test, although alternatives can definitely be used:

1. An iPod Touch (You obviously need a device)
2. EQu EQing software (If you choose a different EQ more accurate the better)
3. Phonak PFE 232s (Use your own headphones to repeat the test)
4. Westone 4R (Use your own headphones to repeat the test)
5. Although I only used 2 pairs of headphones on this test, both showed similar results in what I found out.  I will be more than willing to retest (and probably will in the future) with other IEMs I own.

 
All tests were done in a room with no more than roughly 40 dB of noise heard from my location (used iPods’ Microphone along with 2 SPL loudness apps to test: UE SPL and MetalMed dB).  I use roughly because it’s not 100% accurate, but the loudest noise was my laptop’s fan.
 
Now on to the test.  For the first test, I wanted to see the effects of EQuing exactly 1 frequency (or as close to it) 4+ dB.  In actuality I was EQuing a 100 Hz spike, but also had a few others around it spanning a few 1000 Hz.  I started it at 4; started playing music.  While the music was playing, I turned off the EQ for 10-20 seconds, then turned it back on for 10-20 seconds.  I heard no change.  I then opted to increase it more.  When I hit around 5.5 dB, change was definitely apparent as that frequency started resonating a bit…  However, it wasn’t as if that frequency was EQed 5.5 dB (or 4 when it was that low).  In fact, I did hear small changes in the midrange, but they were just that, miniscule (definitely not even close to 2+ dB).  So the results of this?  EQuing a single frequency doesn’t do much in terms of changing sound.  Obviously the only time this wouldn’t be true is if you were listing to a pure tone (or something close to a pure tone).
 

The Spike I used around 1k.  As you can see, there are some small (< 1 dB increases around there as well)
 
The next test consisted of boosting the high-bass up 1 dB, due to EQu’s software stylings, it created a parametric curve through the entire spectrum.  However, none of these changes were above 1 dB.  This was to test the effects of boosting up 1 dB.  What did I find out?  With the Westones and the Phonaks, the sub-base frequencies were more apparent (bass guitars, etc) although not louder, they just had a better presence.  Some of the midrange also became more audible (lower guitars).  However, these were still minor changes.  In some parts of songs, the change wasn’t audible at all.  In other parts of the same song, it definitely was.  Again, I retested by turning on and off EQ, switching songs, etc.
 

This is the mid-high-bass increase I Used.  The increase between the 128 and 256 Hz is 1 dB.
 
My next test was to create a 1 dB boost in the mid-bass and mid-treble while reducing the mids by 1 dB.  Again, all changes were 1 dB or lower.  This created an EQ that resembled a sine/cosine wave (not exactly, but close).  Changes were much more apparent.  The Midrange definitely did sound much more recessed (as if it was a 2-3 dB drop).  Again, I retested the same way I did before.  These changes were much more audible and this test succeeded in getting change almost all the time.  I really didn’t have to listen for the change, it just hit me.  Again, this was a minor change, but enough to create a loss of detail/clarity along with some veiling. 
 

This is the Sine wive EQ I used, nothing is changed above 1 dB.  That is |change| < 1 dB; Change < (+/-)1 dB
 
So, what are we learning here?  Well, it’s funny how a 4 dB increase (something that should be extremely audible) is almost not even audible compared to the 1 dB increase/decrease sine/cosine wave I created.  Why is this?  Why is it that the same 4 dB increase was less audible than the 1 dB increase in the high-bass (and mid-bass), which was barely audible?  Simply put, our brains hear sounds bundled together…  We hear a wide range of frequencies, not just 1 at a time.  The data is acquired and meshed together (analogy) to create the perceived sound we hear.  The fact is simple, we, as humans, hear sound through the entire frequency range, even little tiny ones, as opposed to one at a time. 
 
Now, I’m going to do one little thing, I’m going to attempt to simulate, to the best of my ability, the changes found in Tyll’s 331 hour burn in of the K701s, to see if it’s audible.  Now, upon looking at the graphs, we see 2 small increases in the treble area ranging around 2.5 dB, while the midrange gains around 1 dB…  The bass changed .5 dB if it were lucky.  So I attempted an EQ simulation of this, to see if it would be audible. Please note that this graph isn't 100% accurate, I tried to approximate the changes I saw to the best of my abilities.  The changes are as follows< I have a .3 dB increase @ 32 Hz, and a .5 dB decrease between 128 and 256 Hz.  There is a 1 dB increase from around 500 Hz to 2000 kHz and replicated spikes around 5 kHz and 10 kHz with 0 dB changes between them.  From around 14 - 16 kHz there is -.4 dB change (if that).
 

This is my attempt at replicating the Frequency Graph K701 hour 331 - K701 hour 0.  It's not 100% accurate, but still shows sign of change.
 
Well, as it turns out, things changed.  Although they are all minor increases here and there, the changes were well apparent.  Also note that lots of minor changes are why we upgrade headphones (if it wasn't for the minor changes 300+ dollar headphones give, we all would be using < 200 dollar headphones).  For starters, the midrange “opened up”.  That is, they became more airy and clear.  Perception wise, things did sound more detailed.  The treble became harsher for me, however, more detail was added in.  A sense of smearing also came about.  Now, the final addition was the increased presence of bass.  This isn’t a quantity increase, it’s actually just a presence increase.  Bass seemed more present, but not louder.  Lower mids also became much clearer as well.  However, the bass and lower-mids were not audible in every song I tested.  In fact, the change was similar to the way the 1 dB increase of the bass (tested above) sounded (some change in this part, but not this, etc). 
 
Comments, questions?  Comment below.  Have your own findings to give us?  Comment below.

 

[khaos974]

Indeed, thanks for confirming the mastering truth that broadband changes are much more audible than narrow band ones.
 
But I think you may have misread Tyll's graphs from innerfidelity: http://www.innerfidelity.com/content/measurement-and-audibility-headphone-break
 

 
How did you get this from the graph above?
 
Quote:

The changes are as follows< I have a .3 dB increase @ 32 Hz, and a .5 dB decrease between 128 and 256 Hz.  There is a 1 dB increase from around 500 Hz to 2000 kHz and replicated spikes around 5 kHz and 10 kHz with 0 dB changes between them.  From around 14 - 16 kHz there is -.4 dB change (if that).

 
 
 

 

[liamstrain]

You were controlling the changes, and so knew what you were changing each time, no? Unless I misunderstand, this is not a blind test in any way and already seriously flawed. 
 
But it is interesting what conclusions you draw from this anyway. 

 

[tinyman392]

Quote:

Indeed, thanks for confirming the mastering truth that broadband changes are much more audible than narrow band ones.
 
But I think you may have misread Tyll's graphs from innerfidelity: http://www.innerfidelity.com/content/measurement-and-audibility-headphone-break
 

 
How did you get this from the graph above?
 
 
 
 

 
I used the animated graph he supplied along with the video he supplied.  That's where I got those changes from.  That's a problem with the data he provides, it isn't uniform throughout, as one graph shows one thing, the other shows another.  Which do we believe, they are both accurate and should be trusted, but we don't know which one is right.  Nor can we assume one is.  However, I did know that the drivers did change about .3-2 dB with his previous posts about the first 65 hours of break in, so I took it that the time lapse should be used.


Quote:

You were controlling the changes, and so knew what you were changing each time, no? Unless I misunderstand, this is not a blind test in any way and already seriously flawed. 
 
But it is interesting what conclusions you draw from this anyway. 

What are you getting at?  What's wrong with an unblind test?

 

[Anaxilus]

Great post!  Hopefully some will gain insight into how we actually perceive music and sound rather than continue to make over-simplified and general universal claims about what is and isn't audible.  The ear and brain are amazing instruments but are also double edged swords that can cut both ways.  If, at the very least, your post can keep people from continuing to throw the baby out with the bath water it will be a job well done.

 

[tinyman392]

OK, quick updates on this, I'm going to try to get a better EQ of the change by comparing the two FR curves using some photo software (to be more accurate).  I can definitely see some inaccuracies in my EQ, however, the treble changes have been double checked to be the right shape (the two spikes like that).  I will redo the mids, only have of the mids go up 1 dB, the rest go up about .2, the bass goes up .2-.5 as well,  I will do a retest tomorrow.

 

[khaos974]

Quote:

 
I used the animated graph he supplied along with the video he supplied.  That's where I got those changes from.  That's a problem with the data he provides, it isn't uniform throughout, as one graph shows one thing, the other shows another.  Which do we believe, they are both accurate and should be trusted, but we don't know which one is right.  Nor can we assume one is.  However, I did know that the drivers did change about .3-2 dB with his previous posts about the first 65 hours of break in, so I took it that the time lapse should be used.

 
You should be careful with this graph, the switch between 0 and 331 isn't smooth, for one thing both graphs aren't exactly at the same scale and there's a slight shift between the two, not to mention that how close the left and right channels are make it difficult to read. Actually, I'd wager that both this graphs and the non moving graph I provided rely on the exact same figures, and the non moving graph is a great deal more accurate at displaying them.

 

[liamstrain]

Quote:

What are you getting at?  What's wrong with an unblind test?

Confirmation bias is seriously problematic when you are both controlling (or aware of) what changes are made, and when you apply them. In short, now you know what to look to hear. Which can skew results pretty seriously - usually increases reported effects. 
 
Same problem with testing cables and such. If you know what was changed, if anything, it affects the results and makes them less reliable as a useful measure. 
 
A well run DBT will go a long way here. 

 

[tinyman392]

Quote:

Confirmation bias is seriously problematic when you are both controlling (or aware of) what changes are made, and when you apply them. In short, now you know what to look to hear. Which can skew results pretty seriously - usually increases reported effects. 
 
Same problem with testing cables and such. If you know what was changed, if anything, it affects the results and makes them less reliable as a useful measure. 
 
A well run DBT will go a long way here. 

So you're claiming psychological artifacts?  Meaning you're blaming the testee, and not the actual observations...  Interesting...  ADding to the problem and hitting it HOme by bringing huManistic problems INto the Equation Man, I'm thinking of a word, but it's not quite coming to me correctly...  It's somewhere in here, I know it. 
 
Like, seriously, you're just going to doubt observations...  I doubt you would have walked up Galileo and stated that all his objects dropped at the same rate because he psychologically wanted it to right?  Don't do it to me.  In the name of science, give a logical reason to explain why this happens, and not doubt the evidence at hand...  Who knows, it might be psychological to not hear it...  I remember saying this somewhere.
 

 

[tinyman392]

Quote:

 
You should be careful with this graph, the switch between 0 and 331 isn't smooth, for one thing both graphs aren't exactly at the same scale and there's a slight shift between the two, not to mention that how close the left and right channels are make it difficult to read. Actually, I'd wager that both this graphs and the non moving graph I provided rely on the exact same figures, and the non moving graph is a great deal more accurate at displaying them.

However, look at the question he's trying to answer, is there evidence that burn in exist.  In other words, does the sound change over time and is it audible...  The only thing you need to prove that the sound changes over time is to have two points where the sound is not the same (I'm using 0 and 331).  To test audibility I'm EQuing (or attempting to).  The graph shifts 1-2 pixels to the left according to Photoshop CS5.  For the approximation of the EQ (I cna't get a perfect EQ, the 1-2 pixels won't make too big of a difference audibly, it'd be like changing a few hundred (at most) frequncies, well, if you look at the single frequency spike I have above, you can see that that wouldn't do too much).  Most of the change is done in the treble, but differentiating between 12100 and 12000 Hz is quite hard.  Even with the 200 Hz differences, the readings will only be 1.25% off... 

 

[khaos974]

Quote:

Quote:

 
You should be careful with this graph, the switch between 0 and 331 isn't smooth, for one thing both graphs aren't exactly at the same scale and there's a slight shift between the two, not to mention that how close the left and right channels are make it difficult to read. Actually, I'd wager that both this graphs and the non moving graph I provided rely on the exact same figures, and the non moving graph is a great deal more accurate at displaying them.

However, look at the question he's trying to answer, is there evidence that burn in exist.  In other words, does the sound change over time and is it audible...  The only thing you need to prove that the sound changes over time is to have two points where the sound is not the same (I'm using 0 and 331).  To test audibility I'm EQuing (or attempting to).  The graph shifts 1-2 pixels to the left according to Photoshop CS5.  For the approximation of the EQ (I cna't get a perfect EQ, the 1-2 pixels won't make too big of a difference audibly, it'd be like changing a few hundred (at most) frequncies, well, if you look at the single frequency spike I have above, you can see that that wouldn't do too much).  Most of the change is done in the treble, but differentiating between 12100 and 12000 Hz is quite hard.  Even with the 200 Hz differences, the readings will only be 1.25% off... 

To me the issue is not about the audibility of the EQ you used, I'm reasonably confident that the EQ you used is audible, even in ABX conditions; what I question is the validity of the data you used to set your EQ, Tyll's second graph:
 

 
is the data set you should be using to set your EQ, it's a more accurate representation of the figures he measured IMHO.
 
 

 

[tinyman392]

@khaos: If you find this post TL;DR, read the last paragraph (and this one), I do go off the audio spectrum a little in this post, but it's used to show why I choose one graph to use over another when both are equally valid.  You are right though, the graph you showed doesn't show any sign of an audible change, but please read on (or at least read the last paragraph).
 
You do make a very good point, however, that graph also contradicts that found here: http://www.innerfidelity.com/content/evidence-headphone-break-page-2  Although it's not a comparison  of 0 to 331, it still is valid and accurate (as accurate) as the second he provided.  However, these also show .5-2 dB changes which can be audible as well (which the other graph is still missing).  To add, it also has less change than the one found at 331 hours which leads to believe in the natural ways of disorder (change away from the original stable state). 
 
Let's look at it this way, there are two ways for the 331 hours to happen:

1. Changes go from A to some point B back to point A again.
2. Changes go from A to some point B to another point C.

 
Now look at the 65 hour graphs Tyll has, now we can get even more of a closer look at it that kind of gives it more direction:

1. With the second graph you provided, it would go: Changes go from point A to point B which is further from A, to point C which is further from A compared to B, to point D which is further from point A compared to point C, etc, etc, back to point A at 331 hours.
2. With the time-lapse graph I use, it would go: Changes go from point A to point B which is further from A, to point C which is further from A compared to B, to point D which is further from point A compared to point C, etc, etc, but eventually to a point Z which is even further from point A compared to point Y. 

 
Now, Although it is entirely possible for it to go full circle, it is more likely to continue changing and stay away from point A because it would defy the idea that the natural progression of things is from ordered (stable) to unordered (changed). 
 
I know this is going out of hand, but it's more natural for things to move away from a stable position than towards a stable position.  This has been well documented in just about every instance possible.  Whether it's from the big scale for the universe since it started in a single, stable position by blew up into an unstable one (which still isn't stable since we are moving away from the big bang's orgin).  When you drop a glass, it shatters into many tiny pieces.  When two atoms collide, things can become unstable.  What you never see is the opposite.  Parts of an atom colliding to form an atom, many pieces of glass coming together to form a glass, etc.  The natural way is change, everything is changing all the time.  I don't deny that that specific graph because it's inaccurate, I agree it has great accuracy.  I deny it because other tests, tests he's done himself show that there should be more change than that that is shown there on two accounts.  Combined with the idea that all things are changing and that's the natural direction.  I can say that although that graph is valid, the time lapse and and earlier graph of the first 65 hours are more valid because they follow the natural patterns of disorder.  Now you may be thinking that I may be pulling ideas out of my a**, but it really isn't.  Even watching an education video like NOVA, they do explain this concept as well (which is my source for this information).
 
I hope my point here makes sense.  Now there still is one question left unanswered, and this question requires a full time lapse of the life of a headphone to see what happens to the actually change in the drivers to be answered.  The question would be, if the way is ordered to unordered, and things are to continue becoming unordered, why does the FR eventually smoothen out (EG, change sounds like it stops)?
 
OK, at this point I've thrown myself off topic, but my reasoning for why I use the timelapse over the non-time lapse is above.  In short, the reason is because the timelapse graphs also follow a similar path compared to that found here: http://www.innerfidelity.com/content/evidence-headphone-break-page-2  Because of that, I find that it's more accurate to use the timelapse because it fits not only one experiment, but two (EG, it's been empirically verified twice).

 

[liamstrain]

Quote:

So you're claiming psychological artifacts?  Meaning you're blaming the testee, and not the actual observations...  Interesting...  ADding to the problem and hitting it HOme by bringing huManistic problems INto the Equation Man, I'm thinking of a word, but it's not quite coming to me correctly...  It's somewhere in here, I know it. 
 

No, I'm blaming the methodology. This is not an ad hominem (which would be me doubting your results because I don't like something about you, or something unrelated to the data), it is quite relevant to the objectivity of the data presented and hardly an obscure request to make. Blind testing (double blind is better) provides more useful data than self controlled sighted testing which is subject to all manner of biases without a control of any sort.
 
If Galileo (or anyone) presents poorly done research, I will feel free to doubt the observations made. "I heard it" is not sufficient to prove that there was something audible when you lack basic controls in your testing. Sorry. 

 

[tinyman392]

Quote:

No, I'm blaming the methodology. This is not an ad hominem (which would be me doubting your results because I don't like something about you, or something unrelated to the data), it is quite relevant to the objectivity of the data presented and hardly an obscure request to make. Blind testing (double blind is better) provides more useful data than self controlled sighted testing which is subject to all manner of biases without a control of any sort.
 
If Galileo (or anyone) presents poorly done research, I will feel free to doubt the observations made. "I heard it" is not sufficient to prove that there was something audible when you lack basic controls in your testing. Sorry. 

Let me expand on the Galileo...  He had a lot of observations that he couldn't prove...  They were just that, observations.  An observation is some piece of data obtained through any of the experimenters five senses.  Although today we can prove his ideas, back then, he couldn't, they couldn't.  Actually, a lot of people called him "crazy" (labeling him) for his ideas that a bowling ball and a marble would drop at the same rate.  They denied that observation(s) (I did do multiple observations) because they believed otherwise.  However, they lacked one thing, empirical evidence to prove him wrong.  He lacked empirical evidence to prove them wrong. 
 
So here's your chance.  Prove your point write by:

1. Proving me wrong
2. Deducting a logical conclusion that can be proven with empirical evidence on why we are hearing differently when you prove me wrong

Some notes, psychological bias is not empirical evidence since it can't be proven empirically unless you own an MRI machine and run brain scans while I'm doing this test.  However, this isn't possible to do since the MRI machine would break because there would be metal in the MRI machine (magnets kinda power this machine; any metals skew results). 
 
I'm fully willing to accept your conclusions if you can prove them empirically.  Otherwise, they will be disregarded as they don't help or attack the evidence, they are mere assumptions on what could be happening.  Could is not accurate enough to prove anything inaccurate.  Again, I'm will to accept your conclusions if you can empirically prove them.  Otherwise, please accept this data as is and not state that it's inaccurate without proof.

 

[khaos974]

The whole part about stability is a confusing mess, you are confusing the concepts of entropy, metastability and stable position of equilibrium, all of which are very different.
Concerning the validity of the graphs you are using, the best would be to ask Tyll about how each graph came to be drawn and that shall put an end to all speculations.