How common is it for VSTs to misbehave when run at 192k?

[househoppin09]

I mostly tend to work at 44.1, but there are times when I'll do a project at 192 if I'm dealing with aliasing issues and need to keep the artifacts to a bare minimum. I haven't noticed any problems so far, but have often heard murmurings here and there about some VSTs showing unexpected or problematic behavior when run at such high sample rates. How common is it for this to be an issue in practice, and can anyone list off any VSTs that are known to have such issues? It would be helpful to know what specifically to watch out for, and just to get a handle on how prevalent this is and how concerned and watchful I should be in general. Thanks!

[DJ Warmonger]

Every VST lists supported sampling rates in specification. Some simply support it, but some do not.

[BertKoor]

If you are afraid of side effects of other plugins (which only is likely to happen when the developer did not do his homework and does his DSP for a fixed sampling frequency) then another more conservative approach is to use Oversampler on the single VST you want to oversample.

[lfm]

Waves have a chart for every plugin they have. You can get a feeling for how common 192k is supported.

Interesting to notice is, when I bought Waves Gold bundle 2013 - there were still three plugins that were 48k only, not even 96k. A year later they fixed this to 96k, not 192k.

So question is really valid.

I'm with BertKoor in this - use oversampler of some sort if certain plugins need it.
I use Metaplugin for this in certain cases for synths. Just click x2 oversampling, and reload project for it to take effect at next load.
Synths in particular render each pitch/note you play so aliasing is a bitch, kind of.

[househoppin09]

Thanks for the tips, everyone. What are the most common symptoms of running a VST at 192k when it's not supported? I don't have any of those Waves plugins, for example--does anyone know what actually happens when you run them at 192k? My worry is that the plugin will continue to work, but in ways that are flawed and compromised, meaning I won't even know there's a problem and will have to waste time hunting down and troubleshooting behaviors that may be difficult to notice at first. Can anyone shed any light on whether that fear is reasonable and what I should do about it? I don't want to run oversamplers on every plugin, nor will I be able to do a perfect job of keeping track of which sample rates are theoretically supported by every single one of the tons and tons of VSTs I use (and that's even assuming that information is available and accurate, which it often won't be).

[lfm]

What happends - if any filters are involved, they may behave all wrong and have crossover frequency at the wrong spot - and what you wanted to avoid, artifacts, will be there to a larger degree. So a simple EQ could sound all different suddenly. Various modulation stuff is likely even more affected.

Many daws have built in channel strip stuff, compressors and EQ etc - and also check if using that at what sample rate they are tested and approved. Just to be sure.

About oversampling - it's not solving anything in itself, just running a plugin at higher sample rate - and meant as solution if just a few in a full project need higher sample rate. So not having to run a full project at 192k, just target some particular plugins and otherwise run project at 96k.

My point with Waves example - maybe theeeeee major plugin vendor out there - and not obvious that everything support 192k. Many stop at 96k, and not so long ago even at 48k.

I have yet to see any equipment that are common that support higher than 96k over ADAT optical as one example. Soundcard converters usually do support 192k, unless usb, but extensions over to preamps I have not seen anything. Already doing 96k - you need double optical ADAT to do 8 channels instead of a single at 48k etc.

It's kind of extreme with 192k. So are you maybe creating more problems than you solve?

[househoppin09]

Hmm, okay. That's exactly what I was afraid of, that crossover freqs, modulation, etc. would be disturbed in ways that might not be entirely obvious at first. Damn. So if I want to run my projects at 192 without introducing any new problems, the only option really is to explicitly confirm support for that with the developers of every single plugin I'm using, and use an undersampling wrapper of some kind on every single plugin that doesn't offer such support? (Come to think of it, do undersampling wrappers even exist? ) I realize that not a lot of hardware supports such high sample rates in any case, but since I do everything ITB that doesn't really concern me.

[lfm]

I read at Steinberg forum at some point people pointed out that even 96k sounded worse on their system.

Why ?
Probably due to more jitter in the samples - the system were working so on the limit on this extra load.
My own interpretation anyway.

And usb is certainly not designed for that kind of precision - so even listening too, and míxing may be affected.
So really careful written drivers for the load at the moment.
Final rendering offline should not introduce jitter, as I understand it.
But to be sure 192k once rendered, I would also convert to more like 48k/44k and see if anything sounds different.

You can always make some measurements doing analog loopback tests with software like RMAA. You will see any artifacts introduced. But I had problems with even generating 96k test pattern, so had to do 48k pattern and use sample rate converter I know are good to get that to 96k(Length became really odd like 10s instead of 1 minute).

But this could give you indication at least, how your system is doing at 48k, 96k or 192k.

I never use Cubase to convert sample rates - after testing this 96k->44k conversions with RMAA(I have not tested 9.5 yet, which has new 64-bit audio engine).
Don't know if this picture with result works - I will try.
Cubase SRC

Comparison Sonar(white), r8brain(cyan), Cubase Pro 9.0(green) and original test pattern in magenta. This is intermodulation(IMD) then.

And don't forget that sample based instruments do their own sample rate conversions in realtime. If you have samples recorded at 44k, which seems most common even these days, it does upsampling/oversampling in realtime for you. so what that introduce in itself you have no idea really. Just look at Cubase in graph and see.

Full samplers like Kontakt, Halion,Independence, you can load you own sample and test like the RMAA test pattern and really calibrate your system so you know what you are dealing with. But readonly sample libraries, you cannot do that - just trust that vendor, or contact them and see what they use. When doing realtime SRC it's always a compromise between performance/cpu and quality. Some daws let you choose what is use in algorithms, like Reaper do. Some samplers I used like Vsampler also let you do that. But Toontrack SuperiorDrummer I have not seen anything(I have SD 2.3, don't know about 3.x).

So to go deep into artifacts, aliasing of various sorts - you might do these once for all tests to see if this is really worth the extra trouble with 192k or not.

[househoppin09]

Wow, that's a lot of info, thanks! I didn't know about RMAA, I'll definitely get a lot of use out of that. Just to be clear, though, I'm not at all worried about the SRC aspect of things. What I'm worried about is the actual DSP behavior of VST instruments and effects themselves. That's obviously not something that can be measured using any tool, as the differences might be much more complex than mere variations in noise, THD, frequency response, etc.

I realize that you're trying to nudge me toward not bothering to run my projects at 192 in the first place--all of the testing you suggest is mostly for the purpose of determining whether it's even worth doing that, right? But I already know it is, because I use a lot of weird old plugins, many of which definitely have major issues with ugly aliasing, which generally clear up at 192 but are often still audible at 96.

So, my only choices here are to run the project at 192, or to run oversampling wrappers on every single plugin I have that gives me aliasing grief, which is a lot of them. Needless to say, I really don't want to have to bother with a dozen or more oversampling wrappers on my larger projects! Therefore, running projects at 192 is the only appealing option, so at this point I'm just trying to figure out how to minimize the downsides and dangers of that. Playback also isn't the issue here--in fact, I don't even actually run my projects at 192 until it's time to render the final audio file. So, while I'm obviously aware of the extreme impracticality of attempting 192k playback on typical hardware, that's not something I'm concerned about.

[Giova942]

Why don't you simply use better plugins?

192K is insane, and you are probably causing more problems when using such a high samplerate.

[murnau]

I mostly tend to work at 44.1, but there are times when I'll do a project at 192 if I'm dealing with aliasing issues and need to keep the artifacts to a bare minimum.

Complete nonsense.

[lfm]

Playback also isn't the issue here--in fact, I don't even actually run my projects at 192 until it's time to render the final audio file. So, while I'm obviously aware of the extreme impracticality of attempting 192k playback on typical hardware, that's not something I'm concerned about.

So you don't listen to what you did and decide more or less of this and that?
I mean, isn't that the core of what mixing is?

You say it makes a difference running at 192k - how?
Yes, you render and then listen, right?

If you actually hear something odd at 96k - it could be jitter and your system is not even up to 96k running in realtime. If usb audio interface I would find it likely.

If the key to it all is what you finally actually hear - then playback must be involved at some point.

Seems to me targeting what you feel are suspected plugin to create too much issues at 96k - is the way to go.

And why stick to plugins that are that old, and create these issues. The more the same sample is resampled to various pitches - the more problems introduced. Look at resample of pitch correcting plugins like Autotune, Melodyne, Waves Tune etc. These involved also moving harmonics correctly - which cannot just be done just resample sample at new rate, much more involved. So if an old synth does a poor job at generating a new note pitch from one sample - increasing to 192k will not correctly move harmonics of that note - if that is what you hear is odd. Aliasing should be diminished though.

Listening to strings from various libraries I have - you feel it sound ok in one octave - but then not as nice in the next octave - so probably due to resampled from one octave to another. Other things are involved as well as formant filtering for that particular instrument and producing other harmonics in another octave.

But actually hearing aliasing at 96k - I wonder?
It could be harmonics that are not where they should be - and you accomplish some kind of filtering that you feel favour running 192k - maybe because the plugins does not really support it. But that it's not actually aliasing anymore that is the issue.

It could be some high shelving LP filter do the same job, actually. I heard that some mixers actually do LP at 10 or 12k - since what you have above that is not significant anyway. It's just children that hear this - so it matters. It's not like - this sounds good because you hear 14-15k - but awful if it's gone. It could be artifacts of notes in lower frequency range even, and everything just benefit from removing it.

Anyway, the discussion is interesting. You often see that certain plugins or hardware now do 384k as "news" - and you wonder - who is this for?
Now we know one guy at least...

[househoppin09]

Okay, I feel like I'm probably still not properly conveying my situation and needs here, which is probably my own fault. Let me try again, I'll paint a brief picture of my approach and hopefully clear up some of the confusion that seems to be arising over what I'm doing and why. I wanted to spare you guys the gory details and keep it brief and simple, but that seems to have backfired...

So, picture this: I'm working on a weird abstract/avant-garde electronic track. It uses lots of common, standard plugins, plus a bunch that are obscure and unusual. The latter would tend to be unique and unconventional sound generators and effects that aren't really replaceable, and in many cases were never even publicly released in the first place. For whatever reason, many of these plugins are written in such a way that they generate ridiculously high-running harmonics and therefore alias noticeably even at 96k, and not intentionally or in a way that I like. I really don't want to waste everyone's time with detailed lists of the problematic plugins or descriptions of what they're doing, so please just believe me when I say that I can't easily replace them with anything more widely available.

So the situation is, I need to use these particular plugins, their aliasing is noticeable even at 96k, and I'm not willing to have it in my final rendered audio file. It also bears repeating that these are pure sound generators and signal processors, nothing sample based and there's no internal sample rate conversion to worry about. That means I can run these plugins at any sample rate I want and not have to worry about any artifacting from sample rate conversion at any point in the process before the downsampling of the final audio render.

Therefore, the solution I've settled on is: I produce the track and roughly mix it at 44.1, simply ignoring and working through the aliasing. I then render my audio file at 192k and downsample that render to 44.1 properly, using high-quality/CPU-intensive dedicated sample rate conversion. The resulting 44.1 audio really is free of any aliasing from my weird sound-generation plugins--this works! Surprisingly enough, relatively little tweaking of the mix is usually required to re-balance things to compensate for the loss of the alias frequencies. At most, I might have to do two or three iterations of tweaking. At the end of it all, I have a render that sounds basically identical to the standard 44.1 playback from within my DAW, but with none of the aliasing on the problematic plugins.

This must all sound insane to some of you, but it's really not that onerous and it gets me exactly the results I need. So far, I haven't run into a single problem. The only other feasible solution, which people have pointed out upthread, would be to use an oversampling wrapper on each problematic plugin one by one. However, that's a pretty unappealing prospect because there are a lot of them and I tend to replace and swap them around quite a bit over the course of developing a track. Without getting too into the weeds on this, let's all just agree to accept that I personally find it much easier to do the "work at 44.1, render at 192, do a high-quality downsample back to 44.1" thing that I described above.

So far, so good, right? I'm getting the results I want and I haven't run into any problems yet. The only reason I even decided to start this thread was because I suddenly realized that, even though it doesn't seem to have happened yet, there's a very real chance that sooner or later I'll throw some mainstream plugin onto one of my projects which, unbeknownst to me, misbehaves at 192. In that case, it could be a nightmare to track down which plugin is misbehaving, much less figure out what to do about it. Some of the posts upthread seem to have confirmed my fears, but I still don't really have a sense of just how common or widespread it is for such misbehavior to actually happen.

Like I said, my process is working out great for my needs so far, so I obviously don't want to start doing something different just because I might run into issues down the line. Unfortunately, it looks like 192k is so rarely used that none of you can really say much from direct experience about what to expect with it, which is perfectly understandable. Still, I'd love to hear any further advice on this that might occur to anyone. Sorry about the length of this post--people seemed to be unclear about what the heck I was up to and I figured I needed to lay it all out in some detail. If there's anything about my process that still doesn't make sense to anyone or anything important that I seem to be overlooking, feel free to harangue me further about it and I'll try to clarify...

[lfm]

Intresting to hear your story.

Without getting too into the weeds on this, let's all just agree to accept that I personally find it much easier to do the "work at 44.1, render at 192, do a high-quality downsample back to 44.1" thing that I described above.

Just a thought, as I saw in Reaper 5.4 I trialed last summer - there is a setting when rendering whether to use project sample rate or final sample rate - when rendering.

I never saw that before on any daw. I always assumed project decide what to render at - and SRC as final operation after bitreduction and dithering.

Using oversampling in a plugin - project at 48k and a synth like Dimension Pro and strings as patch, I could get clear reduction of harmonic beating and other artifacts using 2xoversampling and then as it downsamples again to project sample rate you get a filtering. Estimate bad parts was reduced by 3 dB or so.

But not that obvious if running anything else like rhodes piano and other sound. So depending on complexity of sound - it matters more or less.

And obviously having such an improvement on each component in a mix - that is major in the end and the sound of it all.

If you have not tried oversampling single plugin - just worth a try and you can have one track running original project sample rate and one track oversampled one - and compare both at project rate, if that is 44k.

If doing orchestral range stuff with electronic instruments and target the more complex waveforms that matter more - you may improve your audio quality as you wish and improve quality of you music creating life as well.

[camsr]

Synths take quite a lot of oversampling sometimes. I've found that FM synthesis is the worst, sometimes requiring up to 32x oversampling to relief aliasing in the treble. Like LFM said, it's mostly a matter of the sound's complexity, and how much harmonic energy it generates. More and higher harmonics will end up requiring more oversampling.