Voxengo Premium Membership - All Voxengo Plugins For a Fixed One-Time Fee
Winter Holidays 2017-2018 Discount
Forums     Plugins     CurveEQ EQ matching scale factor


Last

Next

Previous

First



The EQ matching function works well but seems to overcompensate by a huge amount.  For example, where the reference and target curves deviate by 3dB the generated EQ curve seems to compensate by +9dB or more.  It is consistently WAY too much - i.e., it's "over EQ-ing". (At first I thought it didn't work right until I discovered the scale slider to adjust it).

The scale slider works well to pull the EQ curve down to the proper amount - but that amount (scale factor) is not indicated anywhere.  I would like to be more consistent and repeatable in applying these EQ curves.  Is there any way to either:

1) change the default scale factor of the generated EQ curve or,

2) display the amount of scale slider effect (for example, in percentage) or,

3) any other means to adjust the correction curve amount that can be accurately repeated?


It works fine here... (sorry, not a very helpful comment perhaps)
My platform: Win7 home premium x64 // I use Voxengo 32bit & 64bit VST plugs in Reaper 5.x 32bit and 64bit

CurveEQ does not apply any "default" scale factor - it performs "plain" matching which tries to follow differences precisely.  If generated EQ curve compensates by 9dB it means there is 9dB difference exists at certain frequencies which you may be overlooking.

Scale slider does not apply a fixed setting - it is a relative adjustment, so a repeatable adjustment is not possible by design.


Aleksey Vaneev: CurveEQ does not apply any ''default'' scale factor - it performs ''plain'' matching which tries to follow differences precisely.

In that case the "default" scale factor is 1:1 (100%) matching.  That's what I expected but it's not what I'm hearing.  The "matching" EQ that I am getting is massively overcompensating for some reason.  As I stated above, what appears as a 3dB difference between target and reference is generating about a 9dB EQ compensation curve.  If I pull down the scale slider I can MAKE it match but it takes a significant reduction to get there.  Is there any way I can post screen shots here?

Aleksey Vaneev: Scale slider does not apply a fixed setting - it is a relative adjustment, so a repeatable adjustment is not possible by design.

EQ itself is "relative" but we can quantify it with ratios like "dB" - and it is repeatable.  If we pull the scale factor slider down far enough the EQ curve becomes flat.  So there is a quantifiable range between where it starts (100%) and the point where it becomes flat.  Why can't that be displayed as a scale factor?  Seems it could.

These questions are not meant to be negative - I really like this tool.  I'm just trying to find a way to make it more repeatable and quantifiable for mastering purposes.


Aleksey Vaneev: CurveEQ does not apply any ''default'' scale factor - it performs ''plain'' matching which tries to follow differences precisely.

Nonlinear: In that case the ''default'' scale factor is 1:1 (100%) matching.  That's what I expected but it's not what I'm hearing.  The ''matching'' EQ that I am getting is massively overcompensating for some reason.  As I stated above, what appears as a 3dB difference between target and reference is generating about a 9dB EQ compensation curve.  If I pull down the scale slider I can MAKE it match but it takes a significant reduction to get there.  Is there any way I can post screen shots here?

No, forum does not support file posting.  EQ curve may display 9 dB since it is displayed in a centered manner.  But relative to nearby frequencies it may still be 3dB you are seeing.

Aleksey Vaneev: Scale slider does not apply a fixed setting - it is a relative adjustment, so a repeatable adjustment is not possible by design.

Nonlinear: EQ itself is ''relative'' but we can quantify it with ratios like ''dB'' - and it is repeatable.  If we pull the scale factor slider down far enough the EQ curve becomes flat.  So there is a quantifiable range between where it starts (100%) and the point where it becomes flat.  Why can't that be displayed as a scale factor?  Seems it could.

It can't be displayed that way, because equalizer does not hold 2 settings like 100% and flat, the scaling is applied to the current setting.  You can apply scaling several times.


Nonlinear: a 3dB difference between target and reference is generating about a 9dB EQ compensation curve.  If I pull down the scale slider I can MAKE it match but it takes a significant reduction to get there.

May I suggest a test you can do?

Load a sample of a piece of music, or a loop of a song.

Load 2 curve-eqs and no other plugins onto the track

On curve-eq #2, grab the spectrum in slot#1 (remember to set spectrum-display to static and let it stabilize before capture).  It should now show an overlay of that spectrum for reference.

Then on curve-eq #1, set up some curve that alters the sound beyond the 3dBs you mention.

On curve-eq #2, grab this new spectrum in slot#2 (remember to clear the static spectrum-display first, by clicking on it so it resets)

Turn OFF the display of this grabbed spectrum so it isn't visible on the spectrum display.

Now on curve-eq#2 use slot#1 as reference and slot#2 as target, and set it to 60 points, and then click match.

On curve-eq#2 reset the static spectrum-display again but keep it set to static.

Now you should see the grabbed spectrum-display from slot#1 match the 'live' static spectrum-display very closely.

I hope my explanation makes sense :)

Ignore the points on the eq-curve and only look at the actual spectrum.

My platform: Win7 home premium x64 // I use Voxengo 32bit & 64bit VST plugs in Reaper 5.x 32bit and 64bit

Nonlinear: a 3dB difference between target and reference is generating about a 9dB EQ compensation curve.  If I pull down the scale slider I can MAKE it match but it takes a significant reduction to get there.

JELSTUDIO: May I suggest a test you can do?

JELSTUDIO: Load a sample of a piece of music, or a loop of a song.

JELSTUDIO: Load 2 curve-eqs and no other plugins onto the track

JELSTUDIO: On curve-eq #2, grab the spectrum in slot#1 (remember to set spectrum-display to static and let it stabilize before capture).  It should now show an overlay of that spectrum for reference.

JELSTUDIO: Then on curve-eq #1, set up some curve that alters the sound beyond the 3dBs you mention.

JELSTUDIO: On curve-eq #2, grab this new spectrum in slot#2 (remember to clear the static spectrum-display first, by clicking on it so it resets)

JELSTUDIO: Turn OFF the display of this grabbed spectrum so it isn't visible on the spectrum display.

JELSTUDIO: Now on curve-eq#2 use slot#1 as reference and slot#2 as target, and set it to 60 points, and then click match.

JELSTUDIO: On curve-eq#2 reset the static spectrum-display again but keep it set to static.

JELSTUDIO: Now you should see the grabbed spectrum-display from slot#1 match the 'live' static spectrum-display very closely.

JELSTUDIO: I hope my explanation makes sense :)

JELSTUDIO: Ignore the points on the eq-curve and only look at the actual spectrum.

Thank you but I see a flaw here.  Even though you have modified signal #2 with an EQ boost/cut you are comparing two correlated signals.  The amplitudes of the frequency bands may be slightly different but the LOCATIONS of those bands are aligned.  Yes, in that case it should be easy to produce an accurate matching EQ curve.

HOWEVER, when you use EQ matching to compare two DIFFERENT audio files or clips (it's real purpose) there may be localized peaks in one file's spectrum where the other has dips.  So some means of SMOOTHING has to be applied before you can make an appropriate spectrum comparison.  However, what I am hearing and seeing seems to suggest that this algorithm is calculating an exact match at each of the specific EQ points (10, 15, 20, 40, 60, etc.).  If those points occur where the two file's spectrums have localized opposing peaks and dips the resultant "compensation" EQ curve will be way off - which is apparently what I'm hearing.  Some amount of smoothing has to be applied to the spectrums BEFORE the comparison is made (not after).

Now, does the "Smoothing" control in the spectrum display settings have any affect on this (I haven't tried it yet)?  If so, perhaps a 1/3 octave setting would do the trick.  However, if this smoothing setting only affects the display then it will make no difference.


You are correct in that matching may "overreact" if one of the signals severely lacks of certain frequencies.  The "Smoothing" control won't fix that, because it affects only a visual representation, not the actual analysis.

So, the only way to fix this is to try reducing the BlockSize and simultaneously using the lowest number of EQ points possible.

On another note, spectrum matching actually performs the required averaging before the match is made.  It averages the spectrum to the specified number of points.  In other words, everything is in order per your requirement.


Aleksey Vaneev: On another note, spectrum matching actually performs the required averaging before the match is made.  It averages the spectrum to the specified number of points.  In other words, everything is in order per your requirement.

Both target and reference spectrums are smoothed before calculating the difference?  OK, but it appears to me the EQ curve is being smoothed AFTER the points have been calculated.  Otherwise, how do we explain the drastic EQ cuts/boosts that sometimes result when matching two very similar sounds?


Aleksey Vaneev: On another note, spectrum matching actually performs the required averaging before the match is made.  It averages the spectrum to the specified number of points.  In other words, everything is in order per your requirement.

Nonlinear: Both target and reference spectrums are smoothed before calculating the difference?  OK, but it appears to me the EQ curve is being smoothed AFTER the points have been calculated.  Otherwise, how do we explain the drastic EQ cuts/boosts that sometimes result when matching two very similar sounds?

EQ curves definitely get smoothed before calculating the difference.  In fact doing it otherwise would be problematic from the very start.

This topic was last updated 180 days ago, and thus it was archived.  Replying is disabled for this topic.

Last

Next

Previous

First