Anti-Aliasing plugin

[stefancrs]

The easiest way to remove the main audible aliasing
would be HP-filter to cut all frequencies below
played note frequency. E.g. this filter should be dynamic and managed by input midi-notes.

Next step to further remove in-sound aliasing distortion would be comb-filter to
eliminate non-harmonic content of the synthesized note.

--Alex

That would indeed work, but it'd be quite manual. You'd have to specify some kind of frequency window width (so that things don't have to be in perfect tune), you wouldn't be able to handle gliding frequencies etc. And, there IS this plugin for such things:
http://www.acc.umu.se/~bedev/software/vst/#SNIPER

[Lagrange]

stefancrs: I agree that you would have to go manual and that would be impossible in certain circumstances (like a complex solo spanning more than 2 octives eww)..

Thanx for that link for Sniper will check that out today..

C

[stefancrs]

Thank Jonas Norberg, he hangs out here at kvr

[Aleksey Vaneev]

I do not have a technique for this. I just said that this might sound impossible for us now, like removing noise did 20 years ago, but in the future someone might come up with a way. Remember that Alias is not random noise but can actually be described exact mathematically, so when analyzing an audio signal we know exactly where the aliased components will be as long as we know the original sample frequency. The "simple" way would be to dynamically notch out the entire potential alias frequencies; it would maybe not be transparent but it's a start. It should be possible to make such a plugin

Torben

I would be happy if anybody offered me an algorithm to process any saturation algorithm without producing aliasing and thus without using oversampled processing or any special band-limiting. As far as I know, this is fundamentally impossible (otherwise I could apply such aliased saturation and then use this 'magic' aliasing removal process). It is not the same thing as removing spectral components based on their level. In this respect, noise-reducers are also pretty limited. They *can't* hear what is noise and what's not (they can if they use autocorrelation over input signal, but that's only an estimation - in general case, fingerprinting is used, guided by a human person).

Noise is uncertainity after all, and in the case of spectrally-full recording aliasing becomes noise as well, so it can't be judged objectively, and in the case of such recording it's worse, because spectral components aliasing creates are of the same magnitude as desired portion of the signal.