or Connect
Head-Fi.org › Forums › Equipment Forums › Sound Science › How do you EQ a Frequency Response graph to produce a reference sound?
New Posts  All Forums:Forum Nav:

How do you EQ a Frequency Response graph to produce a reference sound?

post #1 of 28
Thread Starter 

Hi,

 

There are a lot of sources for FR graphs, such as Innerfidelity. Once you have the graph, how do you adjust EQ to achieve a more reference type sound?

 

Most of the experts here do it by ear, but for those of us who aren't so talented, can we make ballpark adjustments off the graph, at least as a starting point?

 

Here is a graph of one of my IEMs of interest to use as an example. So looking at the top left FR graph, what is the target curve I'm shooting for for a reference sound and can I simply adjust the EQ by the spread in dBs at each frequency to get there?

 

http://www.innerfidelity.com/images/XiaomiPiston2.pdf

http://www.innerfidelity.com/images/XiaomiPiston2.pdf

 

Any advice would be greatly appreciated. 

 

Thank you.

post #2 of 28

The first thing I would do is even out the line up to 2kHz. For instance, if it's -15dB at 1kHz, make your setting +15dB. Above 2kHz, you don't want to shoot for a flat line. Make it slope down evenly from 0dB 2kHz to -15dB at 6kHz, then have it slope up to about +10dB at 10kHz. Above that, it doesn't matter as much. Just make it taper off to around 0dB from there.

 

You are going to need at the very least a really good 5-10 band parametric equalizer or  15-20 band graphic. The iTunes EQ won't cut it. Also, you will want to EQ subtractively to avoid clipping. Instead of doing +15dB at a certain frequency, pull back all the other frequencies -15dB.

post #3 of 28
Thread Starter 
Quote:
Originally Posted by bigshot View Post
 

The first thing I would do is even out the line up to 2kHz. For instance, if it's -15dB at 1kHz, make your setting +15dB. Above 2kHz, you don't want to shoot for a flat line. Make it slope down evenly from 0dB 2kHz to -15dB at 6kHz, then have it slope up to about +10dB at 10kHz. Above that, it doesn't matter as much. Just make it taper off to around 0dB from there.

 

You are going to need at the very least a really good 5-10 band parametric equalizer or  15-20 band graphic. The iTunes EQ won't cut it. Also, you will want to EQ subtractively to avoid clipping. Instead of doing +15dB at a certain frequency, pull back all the other frequencies -15dB.

 

BS,

 

Thanks for your advice. Is the shape of the curve I'm looking to emulate the same as the Olive-Wetti target (pink line) in this graph from Rin's blog?

 

post #4 of 28

I think that chart is measured and averaged differently than inner fidelity. If you corrected inner fidelity's to match that curve, it would likely be very heavy in the high midrange. I think you'd have to emulate this curve from the bottom end up to 6 or 8kHz, then do a little boost up from 8 to 10.

 

http://www.innerfidelity.com/images/StaxSR009.pdf

 

Oddly enough, the isolation graph looks pretty much like I would like to see the frequency response graph looking.


Edited by bigshot - 7/3/14 at 12:25pm
post #5 of 28

there are several reason why you might want to waste your time to learn doing it for your ears and not just in the general direction of flat.

-it's a great way to learn the relation between frequency and what you're hearing. always very educational to play with EQ.

-in the end you will have to decide if you want the sound to be flat for calibrated machines(electrically flat), or flat for the average human, or flat for olive&welti or another type of compensation, or simply flat for you?

- you never know if the measured headphone has exactly the same response as your own headphone. it's not uncommon to have a few DB differences between left and right, so it wouldn't be surprising to have at least as much between headphones of the same model. + how you wear it +the pads being older ... and you might have another few db of variation.

 

 

 

on that graph from M.R.O's, if you believe olive and welti to have the answer to swag flatness(I'm a big fan personally), then your aim it to EQ with the purpose of making that graph flat.

but as bigshot said, then don't use Tyll measurements at all. it is usually a very bad idea to mix measurements from different sources.

post #6 of 28

All of the response graphs from all the sources on the internet are uncompensated for audible flatness. What you have to do is take a headphone that you know is flat and use that as a model. But that is just a starting point. From there, you have to listen to a lot of good recordings of acoustic instruments and tweak for maximum realism. EQ is a process, not a destination. If it was easy, everyone would do it.

post #7 of 28
Thread Starter 

Thanks so much guys, this is a great tutorial and starting point for me. I took an initial shot based on your tips and it sounds better already. I will try to fine tune from here.

 

Thanks for sharing your wisdom.

post #8 of 28
Quote:
Originally Posted by sithjedi333 View Post
 

 

BS,

 

Thanks for your advice. Is the shape of the curve I'm looking to emulate the same as the Olive-Wetti target (pink line) in this graph from Rin's blog?

 

That's the response of the LEAR BD4.2 with the O-W target subtracted from it. So where the graph deviates from zero, the LEAR deviates from the O-W target.

If Rin had the same type of graph for the Piston 2, you'd just be able to dial in the inverse of the graph into your EQ.

Assuming there was no error from Rin and Olive/Welti using a different dummy head.

 

The actual Olive-Welti target can be found here:

http://www.innerfidelity.com/content/headphone-target-response-curve-research-update

The first graph, black line.

 

What you want to do is match the raw, uncompensated eardrum-level response (the grey line on the Inner Fidelity graph) to the target response. You do this by subtracting the measured level from the target level at each frequency, e.g.:

at 20 hz: The measured level is -30 dB. The target is about 3 dB. Therefore you get +33dB ( 3 - 33 dB)

 

Do this for a wide range of frequencies, and you'll end up with something like this: (just an example, the values are made up):

 

Hz 20 30 40 50 60 70 100 200 300
Level -5 0 0 10 20 27 33 34 36

 

As bigshot said, we want all of these values to be below 0, otherwise we might get clipping at peak levels. So reduce all of your numbers by the highest value, so that it becomes zero and everything else becomes negative. (If you have a preamp slider, you can use that instead to keep levels below 0).

 

Put those numbers into your EQ.

post #9 of 28
Reducing all the EQ sliders by x dB is a pretty bad idea, you won't get an even response that way.
For some reason tyll has chosen to roll off all his compensated responses. It makes no sense.
You should also remember to EQ away any ear canal resonance, mostly heard around 6-8 kHz.
post #10 of 28
Quote:
Originally Posted by higbvuyb View Post
 

Assuming there was no error from Rin and Olive/Welti using a different dummy head

 

I don't think it's a good idea to use a graph made by one person to act as a model to correct a graph made by someone else. I also wouldn't recommend trying to correct for the extreme ends. Pushing 20Hz by 30 dB is not going to get you the results you want. Focus on between 40Hz and 10kHz. Let the rest fall where it may. If you're EQing more than 25Hz or so anywhere in the spectrum, your headphones might not be able to take it. Better to just go a 25dB in that direction and not push things any farther. There's a limit to how much correction inexpensive headphones will take.


Edited by bigshot - 7/6/14 at 12:30pm
post #11 of 28
Quote:
Originally Posted by bigshot View Post
 

 

I don't think it's a good idea to use a graph made by one person to act as a model to correct a graph made by someone else.

Ear simulators are standardised, scientific instruments. The biggest concern here is that Harman's GRAS 43AG may not match the OP's ears. This difference is probably going to dwarf the difference between IF's IEC standard ear simulator and Harman's IEC standard ear simulator.

 

 

Quote:
Originally Posted by bigshot View Post
 

 

I also wouldn't recommend trying to correct for the extreme ends. Pushing 20Hz by 30 dB is not going to get you the results you want.

Have you seen the graph of the headphones in question? It's already ~13dB up (from 1kHz) at 20 Hz. The target is something like 3dB up @ 20 Hz.

The values I used in the example table are example values. They are that large because they are raw values from the graph which is offset by -40 dB so that the raw and compensated graphs do not overlap.

If the OP had headphones which actually were 40dB down somewhere audible, I'd be telling them to buy new headphones.

 

 

 

Quote:
Originally Posted by davidsh View Post

Reducing all the EQ sliders by x dB is a pretty bad idea, you won't get an even response that way.

That's what the slider labelled 'Preamp' is for. Or use a parametric EQ, not a graphical EQ.


Edited by higbvuyb - 7/7/14 at 3:07am
post #12 of 28
Quote:
Originally Posted by higbvuyb View Post
 

Ear simulators are standardised, scientific instruments. The biggest concern here is that Harman's GRAS 43AG may not match the OP's ears. This difference is probably going to dwarf the difference between IF's IEC standard ear simulator and Harman's IEC standard ear simulator.

 

It seems that everyone who measures headphones compensates differently. Two different people's graphs on the same headphones often look different.

post #13 of 28
Quote:
Originally Posted by bigshot View Post
 

 

It seems that everyone who measures headphones compensates differently. Two different people's graphs on the same headphones often look different.

This is certainly true if you use the compensated data (the red/blue lines). However IF also provides the raw, uncompensated data (grey lines).

post #14 of 28

but aren't raw data smoothed and if so, not always with the same margin? I'm asking as a general question, because if you look at goldenears graphs the answer is pretty obvious for them ^_^.

post #15 of 28

Compensation curves and smoothing aside, there can be a difference from headphone placement on the simulator (though this mostly minimized with experienced reviewers), headphone sample-to-sample or production run variance, the difference between simulators (mentioned above; the difference between the simulator and the listener is of even greater concern), and maybe something I'm forgetting at the moment.

 

Regardless, a lot of times you do get very good agreement in results.

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › How do you EQ a Frequency Response graph to produce a reference sound?