This was the only reason why I made the post about this plugin. I couldn't imagine a full session with this single plugin. with only 1 instance.

 

When deciding on the oversampling level, you can't trust that more is automatically better. If you oversample to 32x (unrelated to the bug), many (but not necessarily all) plugins will use the easy way out and repeat the 1x->2x process to get to 2x->4x->8x->16x->32x. In effect, that means inserting two additional filters for every oversampling step. At 32x, that might mean as many as ten or even more filters on top of everything else, all of which introducing their own ringing / rippling - depending on the filter type.

 

+heavliy oversampling causes pre-ringing causing smearing of transients in the low-end.

 

On principle or in this plugin?

 

this^^
pre-ringing if linear phase filters are used and smearing is a general byproduct of oversampling. That's why I keep saying: increase project sample rate instead of using high oversampling factors.

 

Last edited: Sep 10, 2023

I thought this would afford a HP filter.

Right, I know that it can smear a DIRAC but can you (or anyone else) really hear which of these kicks have been oversampled 4x and which 32x?

Spoiler: unnecessarily large window

 

This is a valid argument. However, it is an argument that is out of touch with reality. In a mix, not only one instrument is usually sampled up and down once, but many instruments several times. It’s like with noise or harmonic distortion or aliasing distortion or resonances, etc. It adds up.

 

Last edited: Sep 10, 2023

Transient material like a drum hit is not the best way to determine the effects of oversampling filter. Try it on a full mix, or a harmonically complex sound with a wider and more sustained frequency response, then the differences for in a practical sense 'might' be able to be heard.

The truth is that oversampling implementations vary, and can sometimes have an obvious effect and sometimes the effect is negligible, If the effect is audible, sometimes the effect is desirable, sometimes not. Really just depends on the situation. For example, Fabfilter using Pro-L2's oversampling at x4 tends to sound the best for a HipHop/Trap records that I've worked on, while higher oversampling rates up to 32x had a smoothing effect (ie: a lack of distortion caused by aliasing) in the uppers mids/high frequencies that seemed to work better on R&B and Pop records.

Context is everything.

 

Then you couldn't use large format analogue consoles because noise and saturation would add up.

I was referring to

That's why I tested it with a kick.

 

on principal, the more plugins on your master in high oversampling makes it more noticeable, watch out when mastering per say.

 

Of course it adds up. The smallest changes, made in each channel, can make a huge difference in the overall mix. It's called summing. Why do you think that means you can't use it? You can. But you are comparing apples and oranges again.

Digital does not work like analog. Especially not for non-linear processes. You know that. That's why we use oversampling in the first place. For the reduction of aliasing distortion on the frequency axis, we accept resampling distortion on the time axis. On a mix with 80 tracks, where in 70% of the inserts are at least 1 or 2 (but often more) plugins that generate aliasing distortion, the resampling processes become clearly audible in the result due to the necessary oversampling.

You can test this. Take a mixproject that you mixed at - let's say - 48kHz and set the project sample rate to 96kHz or 192kHz instead. Then change the oversampling factor in all non-linear processors so that the internal samplerate matches that of the plugin in mix 1. Then render both mixes in their project samplerate, downsample them to 44.1kHz and compare them. And then tell me again that you can't hear the difference. You pay the price with more CPU cycles and longer render times, but in return you get a mix with the same amount (sometimes even less) aliasing, less pre-ringing, overall better transient response and better localization.

Of course, it may be that you like the sound of resampling distortion, or that you want to use it creatively. Or you simply don't care. Perfectly fine! But it is measurable and audible.

 

Last edited: Sep 11, 2023

Nah, we're probably talking about different things. It has already been proven that a single clipper in the main out nils with a clipper on all tracks (with the same setting and post fader, ofc). In the same way as 96 channels of a console do not produce a higher noise than one channel because the audio signals sum up too, IYKWIM.
But I was primarily only speaking of the low end transients and I don't know why this shouldn't work with a single kick, because the other sounds (like fem voc or triangle ) don't have that many transients at 60Hz or below.

If I find (time and) plugins of my liking which are capable of 192kz, I'll maybe try this.

 

With all of this complexity, I am pretty happy again that I do not use a lot of plugins that generate harmonics in my mixes...

 

Each channel contributes to the sum. It does not matter if it is the inherent noise or the input signal. What does this have to do with clippers?

Anything below 15Hz is of low interest to me. If there's something that doesn't belong there, the mastering engineer takes care of it.
However, the problem I'm talking about is not in the sub frequency range, but in the highs. And yes, it adds up with each additional oversampled process, in parallel and in series.

Here is an example on Saturn2 - project samplerate is 44.1kHz, 32x OS, drive knob is set to 0% :

Spoiler

EDIT:
Or to make the difference more visible:

Spoiler

The first image is a dirac at 48kHz with 48kHz project samplerate. On the insert I have 3 instances of Saturn2 in series. The drive knob is - again - set to 0% and oversampling is 32x.



In this picture I used the same conditions. The only difference: both the Dirac and the project samplerate is 192kHz.




Both projects were downsampled to 44.1kHz using the r8brain free algo in Reaper.

In terms of transient response, sinc768 would achieve even better results with the 192kHz project. Because less pre-ringing. But I mostly use r8brain because it renders faster. However, here's the difference:

Spoiler

r8brain free:


sinc768:

 

Last edited: Sep 12, 2023

So Kazrog fixed it by ... removing 8x, 16x and 32x oversampling? I don't actually have the plugin, but that seems to be what people say about the update?

 

Spoiler: no, he changed from oversampling factor implementation to adaptive oversampling, so x32 is always 1.536kHz:

 

Last edited: Sep 12, 2023

Yeah, people over at GS are suggesting good = 2x, best = 4x, irrespective of actual sample rate. Can't test though.

 

That's right. We can't because the OP said to wait until the thread ends. Oops... wait a min here... ahem.. is this the delete button ? jh$%&

 

a bit more related to kazrog:
Avalon VT-747SP performs well in terms of oversampling. At least on the time axis.
Same procedure as in the last test. Two dirac for 48kHz and 192kHz project samplerate. The same (oversampling) settings on the avalon. Avalon 3 times in series. At the last step both projects were downsampled to 44.1kHz

Spoiler: Avalon

I did the same for the True Iron for comparison.

Spoiler: True Iron

48kHz - 32x OS


192kHz - 8x OS

 

Last edited: Sep 13, 2023