Newfangled audio has added an "anti-aliasing" filter to Saturate, which, I quote "removes aliasing artifacts better than oversampling"

However, they use oversampling in conjunction with the filter so testing it on its own seems to be impossible.

The first thing I thought of when I saw this was airwindows' UnBox, which uses an averaging filter to essentially leave the high-frequencies undistorted, thus all-but eliminating aliasing.

The method airwindows uses is very good. When distorting, high-end harmonics generally aren't very pleasant anyways, so you get a very nice "fat" sound of the midrange harmonics without any harshness, but afaik this also means that the high frequencies essentially aren't processed at all by the nonlinear algorithm. So, for example, a de-esser or exciter wouldn't be an applicable use for this tech.

I'm not sure what method Newfangled uses, or if it's comparable to airwindows at all.

Invigorate seems to also use anti-aliasing according to their manual, but also has oversampling so testing it out is very difficult. Having oversampling on "low" has a very noticeable, gentle lowpass that seems much softer than your average oversampling filter, but on "max" it looks standard to me.

In image processing, anti-aliasing filters are extremely common nowadays, perhaps surpassing raw oversampling (or MSAA, which is selective oversampling on geometry edges), however in audio, oversampling is far more common.

I'm curious as to any other examples of anti-aliasing "filters" or methods to avoid it that aren't just simply "not processing the high-end", which is what Newfangled audio has supposedly achieved.

 

Here is probably the explanation : you need to oversample BEFORE filtering, because once generated, the aliasing can't be filtered out.
That's why it is not possible to disable oversampling AND use filter.

They oversample, filter over 20 kHz (basically) and downsample to project samplerate.

https://www.reddit.com/r/audioengin...y_does_high_over_sampling_exist_when_you_can/

 

Oversampling almost always has a lowpass filter; I highly doubt by "anti-aliasing filter" they're referring to adding a lowpass filter.
Newfangled Saturate already had oversampling before this, and unless someone can prove that there isn't a lowpass filter in previous versions with oversampling on then that all but confirms they aren't referring to the lowpass filter as the "anti-aliasing" filter.

Moreover, the Newfangled website explicitly differentiates oversampling & "anti-aliasing"

 

I may be wrong, but afaik you don't need to distort high frequencies to get harmonics in the upper range.
You can get extreme aliasing above 15khz with just a sine wave at 20hz if you distort it enough.

But I may be wrong.
Or I may be working for an oversampling company.
I guess we'll never know.

 

I don't use this clipper, but if newfangled actually does bandlimiting without oversampling, then it must be a very (very!) steep filter. Keep in mind that the benefit of anti-aliasing does not come without a cost. You are just exchanging harmonic reflection at nyquist for filter distortion, which affects your impulse and phase response (ringing and smearing). The best antialiasing (in terms of audio integrity) is still working at a higher project sample rate.

 

The logic is that higher frequencies will alias more than lower ones, so you're eliminating most of the problem by leaving the highest frequencies untouched. That's generally how anti-aliasing filters work in image processing and in airwindows' case; working smarter not harder, but overall weaker in eliminating aliasing.

 

Is this true? Or is it just that intermodulation is much more audible in high frequencies?

I'm really asking, I haven't been able to keep this knowledge inside my head even though I watched Dan Worrall's videos a billion times.

 

Another plugin I forgot is Apulsoft's apShaper which uses "antiderivative antialiasing" as well as oversampling. Think this is the key!

https://ccrma.stanford.edu/~jatin/Notebooks/adaa.html
Alt link: https://scribe.rip/@jatinchowdhury1...for-antiderivative-anti-aliasing-d5847167f510


This paper recommends using it in combination with oversampling, which is likely why you only see it in plugins as such. (I'm just gonna assume this is what newfangled audio uses)
I haven't read the whole paper, just skimmed through to make sure this is right. Have no idea how they do it but the results seem good.

 

What do you mean by "leaving the highest frequencies untouched"? At 48kHz sample rate, anything above 4kHz already produces aliasing with f6.

Edit: Ahh... got it. Thanks!

 

Last edited: Aug 24, 2023

this is what we need, not another saturation plugin

 

You beat me to it. I read this article a while ago. Left me with my brain melted but this guy, the famous Chowdurry DSP developer, is a damn genius.

 

It's important to note that this isn't a filter. Anti-aliasing doesn't make use of filters, unlike oversampling - where filtering at least once is required.

ADAA has been around since 2015, but very companies have implemented it (Melda, Newfangled, Chowdhury DSP come to mind).

 

To the tech nerds here, isn't that how downsampling is applied?

That's correct, but then you have to add a lot of harmonics and with high level (to make them audible) and they're all based on the low freq sound part, not the mid and high freqs and this produces a different result.

Ditto.

Does anyone know the reason for this?

 

I'm not sure I get it. Do you mean that, say, if I saturate the hell out of a bass part that it adds tons of harmonics in higher freqs, and then I try to raise the loudness (in whichever way) of the original unsaturated high freqs, it will also raise the added harmonics, thus getting a completely different sound from, say, applying saturation to the overall signal?

Sounds pretty basic and I feel dumb for even asking it lol.

EDIT : Ah ok, I think I understand. Distorting high frequencies adds different harmonics than adding higher harmonics by distorting lower frequencies.

Now I feel even dumber

 

Did you even READ the link i posted ?

"Aliasing, once introduced, exists below Nyquist - the aliasing is reflections in to the audible spectrum. So it can't be removed by filtering, you have to prevent (reduce) it from existing in the first place."

Read this sentence several times until you really understand it.

That's why they oversample first AND filter oversampled signal.
You can't filter aliasing afterward : it is already "folded" into audible Nyquist frequency, along the original signal.
Nothing can filter it out now.

 

Yes, exactly. Importantly, most oversampling methods also use a lowpass in the upsampling stage. Not to actually filter out anything, but to generate interpolated samples.

It's "difficult" if you don't know math. It's hard to overstate just how little original DSP code is used in plugin development - goes for most big name companies too. Everyone borrows from the same 30 year old code by R. B-Johnson, "The Art of VA Filter Design" by V. Zavalishin, or, more recently, the SVF filters that Andy Simper from Cytomic worked out.

Even many of the modes in Valhalla Vintage Verb are lifted from various papers on reverb theory (some linked on Valhalla blog). That's fine though. The Valhalla DSP guy is super clever and has plenty original spins on reverbs.

Ultimately, 93% of plugin developers are really just waiting for somebody else to serve them the math and code on a platter so they can spend more on advertising and GUIs

 

It'd be helpful if you could specify who exactly you're referring to.

 

OP of course

@towerdefense
"Oversampling almost always has a lowpass filter;"

NOT at all
Oversampling consist in using an higher internal samplerate
Nothing to do with filter

Ok i'm out, i already lost too many time

 

Some more insight:
https://www.mdpi.com/2076-3417/10/1/20
https://www.youtube.com/live/eIJRhLLBaik?si=YTebcJ1c_yUIILkV