Convolution effects: increased dB level with increased samplerate (FIXED)

szoroskobold · 02-13-2021, 07:48 AM

tl,dr: goto RESULT

Hello,
Maybe it has already been reported there but I have not found in Bug Reports, so decided to tell.
While using an evaluation version I did an "only stock plugins challenge"; to spare up CPU power, I worked on my mix @48kHz samplerate, then rendered @192kHz in order to have enough frequency headroom for saturation effects (thus to reduce aliasing to non-audible range). The mix sounded just perfect to me. Then I listened to the exported version: the electric guitars became WAY louder, the reverbs sounded exaggerated.
After a long line of experiments of switching plugins one by one, switching samplerate and watching max peak meter for a loop,
:RESULT
I got that
- JS effect 'Convolution Amp/Cab Modeler" made a ~+6dB increase in output by doubling samplerate (48k->96k)
- ReaVerb around ~+3.5 dB (somehow I could not measure the exact value but it was quite audible)
- JS Distortion ~1dB, it was not really audible and even could have been only because of the better waveform reproduction, almost lack of aliasing etc.
:END( of RESULT)

I made measurements with ReaGate, ReaEQ and ReaComp, JS Major Tom Compressor, JS Non-Linear Processor, JS Volume Adjustment, JS Saturation, JS RBJ 4-Band Semi-Parametric EQ as well - none of them had a noticeable effect as listed above.

As this side effect could ruin a whole mix and is uncontrollable (if we do not perfectly know what dB increase to count with each effects, and we may NOT know), I consider this a bug.
I really like to only use the stock plugins for mixing, most of them sound GREAT, they do what they were made for - and not being distracted by shiny fake "knobs", it really makes me increase my creativity. Considering that, I would really appreciate if they worked under any common conditions such as the one I mentioned.
Thank you for understatement.

Tale · 02-14-2021, 07:22 AM

I think you're right about ReaVerb (I haven't tested the others), and I think I know how to fix it: I think ReaVerb does the same as the JSFX convolution code I posted a few years ago, which is scale by sample rate ratio. However, it would seem that this isn't correct; I guess you should scale by IR sum instead.

Justin · 02-14-2021, 10:37 AM

Hi,

What impulses are you noticing this with? Could you share one or two of the most obvious ones that we can test against? Thanks!

szoroskobold · 02-14-2021, 01:22 PM

Sure!
In Convolution Amp/Cab Modeler I was using that Dumble Overdrive Special and Fender Bassman - Fender Bassman ones that come with Reaper; in ReaVerb that one:
https://drive.google.com/file/d/154v...ew?usp=sharing

szoroskobold · 02-14-2021, 03:33 PM

One more thought I came to today.
Until now, I changed the whole project samplerate _before_ rendering. That made that side effect I wrote before.
Now, I tried doing like that: I unchecked the one for resampling (I choose then 512pt) just in the rendering window and rendered the project. After that, neither the guitars did sound louder, nor reverb became exagerated - maybe it sounded a little smoother, probably because of the lack of those unwanted artifacts (I really do not have golden ears, so do not know...). Thus, the problem only occurs while changing the whole project samplerate. That may be needed in some rare cases - of course, better avoid it.

Justin · 02-14-2021, 03:44 PM

Quote:

I worked on my mix @48kHz samplerate, then rendered @192kHz in order to have enough frequency headroom for saturation effects (thus to reduce aliasing to non-audible range). The mix sounded just perfect to me

If it sounded perfect then you should render at 48k!

Anyway, thanks! Latest +dev build has fixes for this (though you'll need to manually adjust the modes of any pre-existing respective plug-ins in order to get the new behavior).

szoroskobold · 02-14-2021, 03:53 PM

Quote:

Originally Posted by Justin

If it sounded perfect then you should render at 48k!

Anyway, thanks! Latest +dev build has fixes for this (though you'll need to manually adjust the modes of any pre-existing respective plug-ins in order to get the new behavior).

Well, when I did today with 512pt Sinc resample, it sounded (adjective as ungramatically as possible) even more perfecteeer to me (-: Maybe was just placebo because I do not really have well-trained ears as I mentioned above, but I ams sure it certainly did not became worse.

Anyway, thank you for your answers and work!

02-13-2021, 07:48 AM	#1
szoroskobold Human being with feelings Join Date: Feb 2021 Posts: 5	Convolution effects: increased dB level with increased samplerate (FIXED) tl,dr: goto RESULT Hello, Maybe it has already been reported there but I have not found in Bug Reports, so decided to tell. While using an evaluation version I did an "only stock plugins challenge"; to spare up CPU power, I worked on my mix @48kHz samplerate, then rendered @192kHz in order to have enough frequency headroom for saturation effects (thus to reduce aliasing to non-audible range). The mix sounded just perfect to me. Then I listened to the exported version: the electric guitars became WAY louder, the reverbs sounded exaggerated. After a long line of experiments of switching plugins one by one, switching samplerate and watching max peak meter for a loop, :RESULT I got that - JS effect 'Convolution Amp/Cab Modeler" made a ~+6dB increase in output by doubling samplerate (48k->96k) - ReaVerb around ~+3.5 dB (somehow I could not measure the exact value but it was quite audible) - JS Distortion ~1dB, it was not really audible and even could have been only because of the better waveform reproduction, almost lack of aliasing etc. :END( of RESULT) I made measurements with ReaGate, ReaEQ and ReaComp, JS Major Tom Compressor, JS Non-Linear Processor, JS Volume Adjustment, JS Saturation, JS RBJ 4-Band Semi-Parametric EQ as well - none of them had a noticeable effect as listed above. As this side effect could ruin a whole mix and is uncontrollable (if we do not perfectly know what dB increase to count with each effects, and we may NOT know), I consider this a bug. I really like to only use the stock plugins for mixing, most of them sound GREAT, they do what they were made for - and not being distracted by shiny fake "knobs", it really makes me increase my creativity. Considering that, I would really appreciate if they worked under any common conditions such as the one I mentioned. Thank you for understatement.

02-14-2021, 07:22 AM	#2
Tale Human being with feelings Join Date: Jul 2008 Location: The Netherlands Posts: 3,753	I think you're right about ReaVerb (I haven't tested the others), and I think I know how to fix it: I think ReaVerb does the same as the JSFX convolution code I posted a few years ago, which is scale by sample rate ratio. However, it would seem that this isn't correct; I guess you should scale by IR sum instead. __________________ Martinic Audio Plugins - Tale's JSFX Pack

02-14-2021, 10:37 AM	#3
Justin Administrator Join Date: Jan 2005 Location: NYC Posts: 16,335	Hi, What impulses are you noticing this with? Could you share one or two of the most obvious ones that we can test against? Thanks!

02-14-2021, 01:22 PM	#4
szoroskobold Human being with feelings Join Date: Feb 2021 Posts: 5	Sure! In Convolution Amp/Cab Modeler I was using that Dumble Overdrive Special and Fender Bassman - Fender Bassman ones that come with Reaper; in ReaVerb that one: https://drive.google.com/file/d/154v...ew?usp=sharing

02-14-2021, 03:33 PM	#5
szoroskobold Human being with feelings Join Date: Feb 2021 Posts: 5	One more thought I came to today. Until now, I changed the whole project samplerate _before_ rendering. That made that side effect I wrote before. Now, I tried doing like that: I unchecked the one for resampling (I choose then 512pt) just in the rendering window and rendered the project. After that, neither the guitars did sound louder, nor reverb became exagerated - maybe it sounded a little smoother, probably because of the lack of those unwanted artifacts (I really do not have golden ears, so do not know...). Thus, the problem only occurs while changing the whole project samplerate. That may be needed in some rare cases - of course, better avoid it.