Is realtime oversampling a dirty/destructive process?

[DaveGamble]

Im waiting to see the example of the distortion plugin capable of producing lots of distortion without aliasing, and without oversampling.

I see a lot of developers in this thread saying its easy to do.

Lets see the plugin.

I don't see lots of developers saying that. And let me clarify my position:

I don't know how to build a distortion plugin capable of producing lots of distortion without either aliasing or oversampling.

The strategy I posted works /fairly/ well, and I've implemented it with limited success (Sonalksis Maxlimit "analogue" mode Clip). It works OK for small amounts of distortion (a few dB of clip), but past that, in comes the aliasing.

But I'm quite sure we'll make advances should it become a hot topic for research.

HOWEVER, the topic of this thread was not originally "Is oversampling the only way to do extreme distortion", which right now, it might fairly be; the topic was much broader, so that's why you're reading lots of comments about the broader issue. This could be the source of the confusion.

Dave.

[kvaca]

It's recommended to actually READ, what other people write before engaging in a discussion/conversation.

Im sorry,I thought we are in effects forum talking about distortion and aliasing..
but obviously not,becouse now we are in instruments talking about addictive synths...my bad

[Gamma-UT]

No oversampling

4x Oversampling of all compressors


Nobody commented on these. Am I imagining things or can somebody hear the annoyance thing happening at the upper mid frequencies on the non-oversampled one? Would be very interesting to hear your thoughts.

Cheers!
bManic

THANKS for making this example. It is strange more people have not commented on it.

There is a perceptible difference in levels, which means dragging it into a DAW and adjusting for that before being able to make a comment. And, what's obvious in a mix you've worked on versus someone who is hearing it for the first time is not straightforward. I'm not convinced I've heard what bmanic hears but there is a distinct crunchiness or harshness to the percussion. It could be aliasing - I've heard similar effects when moving a project from 44.1k to something higher. But, in these situations, it could just be the plugin reacting differently because of the subtle spread in energy when operating at with a higher Nyquist - hence the "higher is quieter" comment a few posts above.

[bmanic]

Harmor can do that by using the "Prism" feature.
Sounds "spacy" in my opinion.

but Harmor is NOT a distortion plugin, its a synth -isnt it?

It is. But with a synth you can get a general idea of how inharmonic content affects the harmonic content. That's what I'm interested in.

@Nokenoku: Thanks for the suggestion! I do indeed own harmor so I'll need to investigate.

Cheers!
bManic

[bmanic]

The oversampled sounds much better on my laptop, BUT i know from experience i don't like oversampled (up then down to 44.1) sound on my proper speakers and DAC. Its subtle, but the sound gets smaller and weaker. I am sure it depends on what software is doing the oversampling and how, but i can't think of any software I have tried with user selectable oversample rate where it doesn't happen.

Awesome! That makes two of us!!

I really dislike what up/down sampling does to sounds, especially drums. But it is an unfortunate necessary evil. This is why I now do all my mixing at 96kHz. Thsi way I can use R8brain pro or similar high quality oversampling to do the up/down sampling only once for each track instead of running through multiple plugins own better/worse up-down sampling process.

For some very weird reason I seem to hate linear-phase oversampling way more than minimum phase filtering. However, I'm not entirely sure if this is purely because of a current state of mind.. aka placebo. I haven't done any proper ABX testing for this yet and I admit that the differences are very subtle.

The weird thing is that linear-phase oversampling seems a bit constrained.. or weak in a weird way, pretty much exactly as you describe it! The bad part is that it also improves a lot of plugins, especially those that allow for quite a high number of oversampling. So it's a trade off. Do I oversample to minimize aliasing/inharmonic distortion or do I keep it non-oversampled to maintain the "punch"?

NOTE: in my example I'm running my own compressor algorithm and it is VERY possible that the implementation is buggy and thus the "improvement" you hear might not be at all due to less aliasing (although there is significantly less aliasing at 4x oversampling. This is easily measurable). It also makes the side-chain, which is excruciatingly complex and program dependent, much more accurate and thus affects the compression response to quite a large degree. So the improvements are perhaps not at all due to less aliasing.

It is possible that this is the case with The Glue as well. When the accuracy of the side-chain enhanced the compression action will sound more fluid and thus the compressed signals sit better in the mix.. they are less "chunky" or "nervous".

So many variables and possibilities. Wish I had studied more maths and programming when I was younger. My lack of knowledge in these fields are haunting me every day now.

Cheers!
bManic

[bmanic]

It is strange more people have not commented on it.

What's to comment? It's a test for a specific plugin - no one argues about differences, but the reason behind them.

You can render a synth & have less aliasing when oversampling it. Is that a proof that a synth needs oversampling to have less aliasing? No. It's just a proof that one synth does.

Very true. This is why I should have been more specific: What do you guys HEAR as the improvement/deterioration? To me it's very specifically the 2-5kHz area which feels constrained and annoying.. even "messy" in the non-oversampled version. I can not play it as loud on my headphones or speakers as I can the oversampled version, before ear fatigue hits.

Cheers!
bManic

[DaveGamble]

@bManic - Have you listened to the phase cancelled signal of one minus the other?

It's about 20dB below peak, so it's occasionally audible, but what's clear is that the two versions are computing gain reduction differently - basically, the phase cancel is mostly pops, clicks, and bits of the start of things.

Interestingly, there's little high end in the phase cancel - what's there looks more like a dither shape.

Have a play with that. It's interesting. Towards the end of the phase cancel there's a bit where it generates some noise - intriguing. I'd investigate that.

Dave.

[bmanic]

Yes I've done this.. and I don't really know what happens. I suspect the oversampling is messing with the various smoothing filters I have within the gain reduction/VCA model thing which in turn makes the compressor behave "chunkier" or smoother.

When sending some short pulses (just a few milliseconds) I get quite interesting behavior in the way the compressor works when oversampled or not oversampled. Will have to investigate.

I do get similar results with The Glue. There is a clear change in transients there as well.. depending on attack and release settings it seems.

This is what intrigues me a lot. Perhaps feeding the sidechain with a massively oversampled signal simply gives it so much more room to work? Envelope detection has a lot more to work with etc. Or perhaps the magic happens in the VCA / gain reduction element? So much to learn..

EDIT: I checked the compressor at no-oversampling versus 4x oversampling without having any gain reduction happening and as expected, there were almost no differences in the files (difference almost down to noisefloor).. only a tiny difference due to the oversampling filters phase changes at high frequencies (because I sample the whole audio signal, not the side-chain only). So it must have something to do with my implementation.

EDIT2: Compassion also has this behavior but to a much tinier degree. I suspect the difference would be larger if it allowed higher oversampling amounts. On The Glue it seems to get proportionally larger the more you oversample. At least the difference on one test track is 12dB between the peaks when comparing no oversampling to 16x oversampling.

EDIT3: Any synthmaker gurus here? Basically what I'm using is the oversampling toolkit (http://www.synthmaker.co.uk/dokuwiki/do ... ersampling) for my oversampling needs but I'm starting to suspect that this might partly be one of my problems. Any more elegant ways to do the oversampling over a large block in Synth Maker? What about oversampling in Synth Edit? Anybody created an oversampled container so that I could build my compressor ideas in there instead? I'm way too busy to start learning C++ and Juice.

Cheers!
bManic

[bmanic]

Towards the end of the phase cancel there's a bit where it generates some noise - intriguing. I'd investigate that.

Dave.

Duh.. now I understand what's happening in the example I posted. It's not just my compressor which is oversampled. In both examples I also use 2 different limiters in series for the final volume. In the first example none of the effects in the track are oversampled. In the second example everything, including the limiters (demo version of A.O.M Invisible Limiter feeding into Pro-L) are oversampled to their maximum values, AOM limiter going all the way to 16x oversampling. So this probably explains the dithering stuff as I naturally had dither on in Pro-L. It's possible that I had selected two different dither types. I didn't pay too much attention to that.

So the whole chain in those examples is:

Pro-Q
My compressor x3 (in a multiband configuration.. band splitting done within Reaper using it's own plugins)
Another instance of my compressor, this time on the master.. not multiband
A.O.M Invisble Limiter demo
Pro-L

The weird "transient" chunk residue is mainly due to my compressor though. Both limiters do have a bit of transient residue as well but not nearly as much as my compressor so I guess I'm back to square one.

Cheers!
bManic

[@midnight]

Yall are killin my brains!

[mandolarian]

have a bit of transient residue...

Is this thread turning into a shampoo commercial?

(Disclaimer: I did like the non-over-shampooed version better, which is probably not something to mention in the middle of your shampoo commercial. Sorry.)

[meloco_go]

@bManic - Have you listened to the phase cancelled signal of one minus the other?

It's about 20dB below peak, so it's occasionally audible, but what's clear is that the two versions are computing gain reduction differently - basically, the phase cancel is mostly pops, clicks, and bits of the start of things.

Interestingly, there's little high end in the phase cancel - what's there looks more like a dither shape.

Those pops and clicks are definitely results of compressor acting at different rates.
In my tests I used basic non-oversampled compressor and used it at various sample-rates and even compensating for filters there's that click and pop sound on transient.
I think it is clearly interaction of sidechain with intersample information.
When compressor operates on discreet signals it has only a limited info on a signal, the higher the rate, the more correctly sidechain would operate.

[bmanic]

Indeed. This seems to be the case and confirms what is heard.

Cheers!
bManic