4X even 8X.... hell 16X Oversampling (Upsampling)

[LittleStudios]

I know that there's currently built in 2X upsampling on plugins that can be enabled, and that's awesome.  I don't know of any other DAWs that can do that.  I say, take that to the next level!! 4X... 8X.... 16X !!! Computers today can handle those types of computations.  Aliasing causes some real nasty sounds to come out of the speakers. Having the ability to oversample (upsample) the plugins to these levels would put Cakewalk head and shoulders above the rest (at least to my knowledge).  I prefer upsampling my plugins over setting a high sample rate for the project.  Upsampling plugins instead of a high sample rate on the project keeps audio file sizes wayyyyyyy down, keep the audio quality as good as digital can get it.  C'mon Noel, make it happen!!!!

[scook]

The 2x button does not mean upsampling is limited to 2x the current project sample rate.

This is the default.

While sample rates greater than 2x the project sample rate are unlikely to be beneficial in most cases, the actual upsample rate may be set per-plug-in.

The process is documented here.

I wrote a graphical front end to the upsample section of aud.ini called Upsample Editor. It is part of my CbB Tools available from a link in this thread.

[LittleStudios]

Scook.... YOU RULE!!!

 

[Starship Krupa]

It seems like a sudden rash of people clamoring for Cakewalk to help with some angst they are experiencing around the possibility that unless the DAW allows for upsampling plug-ins, their productions will be plagued with aliasing.

As far as I know, I have never experienced "nasty sounds coming out of the speakers" due to a plug-in being insufficiently sampled. Intersample clipping, yes, indeed, but not this aliasing they speak of. Wouldn't it result in certain plug-ins being less popular? Why wouldn't plug-in manufacturers build it in to their own products?

Edited October 3, 2021 by Starship Krupa

[LittleStudios]

I'm not sure if what I've stumbled upon is expected behavior from the software or not.  In testing the upsample feature I decided to take a kick drum track and use it to trigger a noise gate on another track.  The other track had pink noise on it just for testing purposes.  The plugin upsampling was enabled on the noise gate and the upsampling was left on default.  When upsampling was disabled everything functioned as expected.  As soon as upsampling was enabled the sample rate of the audio being sent to the gate and the resulting audio being let through the gate was very wrong.

Here's a link to a video example I put together.  You will have to pause the video to read the captions.

Here's the link: https://youtu.be/CfMU93SCigU

So far, other than this odd behavior, everything is working wonderfully.

 

Edited October 3, 2021 by LittleStudios

[Starship Krupa]

2 hours ago, LittleStudios said:

The plugin upsampling was enabled on the noise gate and the upsampling was left on default. 

Why would you enable oversampling for a gate? A gate is a utility that is either in a state of passing or not passing audio, otherwise in the most linear way possible.

2 hours ago, LittleStudios said:

When upsampling was disabled everything functioned as expected.

Which is one of the reasons oversampling is optional: it sometimes makes plug-ins do weird things, and that can even include weird things that you can hear and see on an analyzer.

You seem to be so concerned with this, how much have you read up on it (and I don't mean amateur opinions on web forums)?

From Meldaproduction's documentation:

"Oversampling can potentially improve sound quality by processing at a higher sample rate. Processors such as compressors, saturators, distortions etc., which employ nonlinear processing generate higher harmonics of the existing frequencies. If these frequencies exceed the Nyquist rate, which equals half of the sampling rate, they get mirrored back under the Nyquist rate."

Got that? If you're using a processor that happens to generate frequencies above 22KHz (assuming you're recording and mixing at 44.1K), then you will potentially get harmonics back down below 22KHz.

Also:

"Finally, and most importantly, oversampling creates some artifacts of its own and for some algorithms processing at higher sampling rates can actually lower the audio quality, or at least change the sound character. Your ears should always be the final judge.

As always, use this feature ONLY if you can actually hear the difference. It is a common misconception that oversampling is a miraculous cure all that makes your audio sound better. That is absolutely not the case. Ideally, you should work in a higher sampling rate (96kHz is almost always enough), while limiting the use of oversampling to some heavily distorting processors."

Got that? Mr. Meldaproduction says that the ideal practice is to record and mix your whole project at a higher rate if you really want to eliminate it. Most importantly, he suggests only using oversampling if you can hear the difference and prefer the sound with the oversampling.

From Sonarworks' site:

"Oversampling benefits the kinds of plugins that change the shape of the original waveform or create new frequency content....Plugins that benefit from oversampling include compressors, limiter, clippers, amp simulators, saturators, and exciters, but not usually equalizers or time-based processors, unless they also provide some kind of saturation."

Moreover, the people who code plug-ins tend to know their stuff, and they're not helpless to prevent anti-aliasing in the plug-ins' internal algorithms. They can put in filters that contain the overtones to the safe range. When they are coding their processors, they expect that they're going to be used normally, which means at the same sampling rate that the rest of the project uses. If we use them outside their design parameters, results are not 100% predictable. Ideally, external oversampling makes it easier to stay away from the Nyquist frequency and therefore avoid the possibility of aliasing artifacts, but that's not guaranteed. There could be some filter or other process in the code that behaves differently, even in an undesirable way, when the plug-in is presented with oversampling.

A frequency response curve might change, level might change (I had this happen with a compressor plug-in).

Speaking for myself, I think I'd do better to put the effort into working on my mic placement than trying to figure out oversampling my plug-ins.

[LittleStudios]

@Starship Krupa You seem to have a very strong opinion about whether or not people should be using oversampling or not.  Also you seem to have a very strong opinion on whether or not aliasing is audible.  That's fine, everyone is entitled to their opinion.  I'm not sure that my feature suggestion of higher oversampling on the plugins (which apparently already exists) warranted this response.  Just the same, to answer your question...

1 hour ago, Starship Krupa said:

Why would you enable oversampling for a gate? A gate is a utility that is either in a state of passing or not passing audio, otherwise in the most linear way possible.

I typically use oversampling only when I feel it's needed.  Mostly on saturation plugins or analog emulations.  In a real world situation, I would not oversample a linear noise gate or any plugin that does not produce harmonics, saturation or any type of analog non-linearity.  What would be the point?  I wasn't originally using a noise gate plugin.  I was originally using a channel strip plugin that had a noise gate (Scheps Omni Channel).  The Scheps Omni Channel has the ability to produce harmonics and saturation.  It also produces audible aliasing.   The basic noise gate was used to simplify things.  To help illustrate what was happening.  

The point of my post was to ask whether or not this was the intended behavior. 

7 hours ago, LittleStudios said:

I'm not sure if what I've stumbled upon is expected behavior from the software or not.  In testing the upsample feature I decided to take a kick drum track and use it to trigger a noise gate on another track.  The other track had pink noise on it just for testing purposes.  The plugin upsampling was enabled on the noise gate and the upsampling was left on default.  When upsampling was disabled everything functioned as expected.  As soon as upsampling was enabled the sample rate of the audio being sent to the gate and the resulting audio being let through the gate was very wrong.

I'm not going to get into a debate on whether high project sample rates make an audible difference.  Or whether oversampling plugins is worth it.  I thought this area of the forum was for providing feedback and feature requests.  I didn't realize that it was a place to also have people tell you that you are wrong for even providing such feedback or feature requests.

I've provided feedback and a feature request.  That's all I came here for.

 

[Starship Krupa]

1 hour ago, LittleStudios said:

You seem to have a very strong opinion about whether or not people should be using oversampling or not.

No, not at all, Chris. And I certainly apologize if I gave the impression that I thought you were "wrong" about anything. You've definitely experimented more with plug-in oversampling than I have.

I come here for discussions, and I usually assume that "why are you seeking that feature?" when I don't understand, is a fair matter for discussion. I've learned some things from people that way. I understand that it can come off as "why the heck would anyone want to do that??"

You said in your first post that "Upsampling plugins instead of a high sample rate on the project keeps audio file sizes wayyyyyyy down, keep the audio quality as good as digital can get it," which sounded simplistic to me. There are many things that influence keeping "audio quality as good as digital can get it," and from what I've read elsewhere, plug-in oversampling is a pretty minor one. YMMV.

I don't really know  much about this topic except what I've read in articles that all take pains to warn against seeing plug-in oversampling as a guaranteed positive thing and what I've heard switching it off and on myself. Which was nothing, but then I don't use a lot of guitar amp sims. Supposedly one of the best tests of whether a distortion plug-in is aliasing is to play a double stop and bend one of the notes. Never tried it. That's why I quoted articles and asked questions. I wasn't coming from a position of knowing it all, rather the opposite. Also, my ears turned 60 this year, and I played in rock bands for about 1/3 of that time, so if the aliasing is up above 15KHz, I may be missing it entirely.

However, I'm one of those people who must set his computer's music playback up to have as close to bit-perfect reproduction as possible. I can hear the difference between CD audio played back via ASIO or WASAPI and via Direct Sound, and it's a big one. The former sounds "3-D" and has "depth" to me, while the latter sounds "blurred" and "flat" by comparison.

The night that I first set my system up that way I stayed up until dawn listening to my favorite albums because I was hearing so much stuff in them I never had before. This effect is as apparent to me on the onboard sound in my laptop as it is on the interfaces in my studio, so I'm always watching out for this "blurring" being introduced by software I use and want to know about any tips for preventing it. I only wish recording and mixing at 88 or 96 made a difference I could hear, I'd do it every time. As it is, CD quality is plenty as long as it's reproduced correctly, without further conversion.

I have witnessed the effects of aliasing via a spectrum analyzer while QA testing DAW software. I ran some tests of sine sweeps through a DAW's (not Cakewalk) sample rate conversions and found that a couple of the permutations resulted in visible (and in the audible range, 10K area) artifacts. I was inspired to do so by this page: https://src.infinitewave.ca/ If you're inclined, you may find it as fascinating as I did. And notice that SONAR, as it was when they tested it, gave some of the most crystal clear results.

I have had one effect, ADHD Leveling Tool, a very nice LA/3Aish compressor, mess up badly when Cakewalk's 2X and 64-bit engine were both engaged. Its output level dropped way down. This suggests to me that there may be other, possibly negative, side effects to oversampling certain plug-ins, and I don't know which plug-ins or side effects they would be, so I keep cautious.

I turn on the plug-ins' internal 2X oversampling for my Meldaproduction fx during rendering only, because what the heck, it's free, and I trust Vojtech to code the function correctly. I listen for unwanted effects and don't hear any, so it's all good. I have to confess that I don't hear any positive effect either, though.

If anyone else wants to let it rip with 16X oversampling all their plug-ins, it ain't up to me to tell them not to, but I myself approach it with some caution and wanted to say so. The possibility exists, at least in my mind, that feeding audio into these things at the rate they're expecting might be best practice, at least for some of them. I'm not sure though. I doubt the Cakewalk engineers would create such a feature and make it that easily accessible if there were too many possible drawbacks.

By all means, carry on, I'm glad that you had the experience, as I have many times, of finding out that the feature you were looking for already exists in Cakewalk. And if you can share more specific experiences with plug-in oversampling, I'd love to hear about them.

[Starship Krupa]

I'm working on my 10-year-old laptop, and, inspired by this thread, decided to use it to see what performance hit I might see from enabling upsampling for the plug-ins in this project.

It uses 4 synths and 6 FX, and the laptop can play it back fine with no upsampling engaged. Onboard audio CODEC, 736 samples. 8G RAM, i7 760 processor. Kinda old by current standards, but I keep it optimized and so far it has refused to fail to perform whatever tasks I've thrown at it, including DAW and video NLE work. It has a discrete nVidia GPU.

Results:

With no upsampling, engine load hovers around 45%, with a spike up to 62%. With all 10 plug-ins set to upsample 2X, it hovers around 80%, with multiple late buffers, stutters and usage spikes up to 146%. Clearly, this suggests there is some expense involved in upsampling.

One of the synths, however, is A|A|S Player running a String Studio patch that arpeggiates. A|A|S's engines, while they sound amazing, are also the most demanding I run on my system. I limit them to 8 simultaneous voices so that they won't bring the show down. Turning it off for just that synth restored gapless playback with spikes up to 82%. Also turning it off for iZotope Exponential Phoenix Stereo Reverb brought the spike down to 72%. I was just guessing at what would be the most expensive plug-ins.

Interestingly, switching the sampling rate in Preferences to 88.2KHz resulted in smooth playback, with usage spiking up to the high 90's. Barely viable, but less expensive than per-plug-in upsampling. This suggests that the tradeoff will be as Chris says, the amount of room that recorded and rendered audio will take up on the hard drive. If space on your SSD is dear, but you have a faster, modern processor with many cores, per-plug-in may be the answer. Since for me it's the other way (as it is for me, I just swapped my DVD+R drive for a second SSD on the portable and don't need to archive projects on it anyway), if I wanted the potential benefits, I'd run at 88.2. Rendering at that rate would mean an extra rate change before distribution, though.

I'll try the same project on the main DAW and see what it says. It may just be that the upsample button on the laptop will have to stay off. I always do final renders on the main system anyway. Jury's still out on whether I hear any differences, but I will do further tests, including with headphones.

Thank you, @LittleStudios for bringing this up. I've long wondered what the performance hit might be.

[Teegarden]

What I still don't get from the manual:

To globally enable/disable upsampling for a plug-in, click the FX icon in the upper left corner of a plug-in window, and select
Upsample on Render or Upsample on Playback on the drop-down menu. These options globally persists for all instances of the
plug-in in all projects, so it only needs to be set once per plug-in.

54 minutes ago, Starship Krupa said:

With no upsampling, engine load hovers around 45%, with a spike up to 62%. With all 10 plug-ins set to upsample 2X, it hovers around 80%, with multiple late buffers, stutters and usage spikes up to 146%. Clearly, this suggests there is some expense involved in upsampling.

If you've set it to global, it doesn't mean that it is active on all plugins that support it automatically? You still need to activate it per plugin first?

If not, keeping it on globally means a serious hit on overall performance according to your results (if you didn't realise that you need to (de)activate per plugin)...

Interesting result, since the manual implies that it should be less heavy on the system:

"While you can work around these problems by using higher project sample rates like 88.2 kHz or 96 kHz, doing so can also add CPU load to the project due to the higher data bandwidth."

 

54 minutes ago, Starship Krupa said:

Interestingly, switching the sampling rate in Preferences to 88.2KHz resulted in smooth playback, with usage spiking up to the high 90's. Barely viable, but less expensive than per-plug-in upsampling. This suggests that the tradeoff will be as Chris says, the amount of room that recorded and rendered audio will take up on the hard drive. If space on your SSD is dear, but you have a faster, modern processor with many cores, per-plug-in may be the answer. Since for me it's the other way (as it is for me, I just swapped my DVD+R drive for a second SSD on the portable and don't need to archive projects on it anyway), if I wanted the potential benefits, I'd run at 88.2. Rendering at that rate would mean an extra rate change before distribution, though.

So, according to your results when using  my standard 96kHz setting, sound quality and performance should be even better without the 2x setting... Or, does de 2x oversampling still double the sampling of the 88.2 or 96 kHz settings (resulting in even more performance hits)? From the manual:

"Cakewalk provides another solution, which lets you specify whether a VST or DirectX plug-in effect or instrument should be
resampled at 2x the project sample rate when bouncing, rendering, freezing, exporting, etc., or during playback."

And if this is the case, would it be useful to still use the additional oversampling for certain plugins (resulting in at least 4 times oversampling)?

I've still got the feeling I'm missing something here...

Edited October 4, 2021 by Teegarden

[LittleStudios]

@Teegarden  You first need to tell Cakewalk which plugins you want to have the oversampling enabled on, by accessing the "FX" icon in the top left of the plugin editor.

Figure 1:

Once you enable this plugin to use oversampling, you then have to be sure to globally enable plugin oversampling by clicking this icon:

Figure 2:

From my tests, if you don't first enable oversampling (upsampling) on a per plugin basis (Figure 1), enabling the global oversampling (Figure 2) will not do anything.  To rephrase, plugins must have oversampling enabled for plugin oversampling to take effect.  If the plugin has it's own built in oversampling, it will act as another multiplier depending on whatever multiplier the developer coded into the plugin, if you've enabled plugin oversampling on both the per plugin setting and global oversampling bypass/enable button.

The default value of Cakewalk's per plugin oversampling is twice the sample rate of the project up to 384,000 Hz.  You can manually input the oversampling sample rate in the AUD.INI file or use Scook's tool linked above in Scook's post.

 

 

 

[Teegarden]

@LittleStudios, thanks for the useful info! 

Question is: does using higher project sample rates like 88.2 kHz or 96 kHz not give the same results or better as 44.1 kHz with global sampling (2x) + activated plugin? And does a 96kHz project with the global sampling (2x, -or 4x, 6x,... times) + activated plugin again provide better results?

Of course, there's always a trade-off between high quality settings and system resources, so where to draw the line? Any particular plugin that would specifically benefit from over-the-top settings (look-ahead plugins, IR-reverbs, Scheps Omni Channel, any other VST)?

[Starship Krupa]

2 hours ago, Teegarden said:

@LittleStudios, thanks for the useful info! 

Question is: does using higher project sample rates like 88.2 kHz or 96 kHz not give the same results or better as 44.1 kHz with global sampling (2x) + activated plugin? And does a 96kHz project with the global sampling (2x, -or 4x, 6x,... times) + activated plugin again provide better results?

Of course, there's always a trade-off between high quality settings and system resources, so where to draw the line? Any particular plugin that would specifically benefit from over-the-top settings (look-ahead plugins, IR-reverbs, Scheps Omni Channel, any other VST)?

Maybe my posts were too verbose, but if you scroll up a couple, I quoted a couple of companies' take on what fx may benefit. It seems that the more a plug-in might generate harmonics above 22KHz, the greater the chance that this information, while inaudible to humans, will get aliased back down into the audible range. If Omni Channel generates saturation or other distortion, then it might potentially benefit.

Vojtech Meluzin (aka Meldaproduction) states the opinion that running the whole engine at 88.2 or higher is the best practice. But he also says that listening to the results is the way to know.

@Noel Borthwick mentions that oversampling (and presumably also running at a higher rate) may help with phase shift, which very much interests me. He also mentions reverbs as being a type that may benefit.

If you take a look at the results of the tests I ran, I came to the conclusion that if you want to experiment with the possible benefits of oversampling, running at 88.2 incurred a significantly lower performance hit than enabling all the plug-ins. But the plug-in that caused the biggest hit (no pun intended) was a synth, not a processor. My Plugin Alliance elysia mpressor, alpha compressor, and Millenia NSEQ caused much less performance hits. AIR Hybrid, much less of a hit than A|A|S Player.

A thing to remember is that all of this is only important during mixing and rendering, so if you crank up the latency in your driver settings, any performance hit will be less of an issue. Latency is usually only an issue during overdubbing, and recording soft synths, so during that process, you can hit the 2X button or just bypass your performance-hungry FX. I deliberately dropped my usual mixing latency on the laptop to make the results more obvious.

One of the things that led to confusion when I first started using Cakewalk was its use of the term "global." I've always understood that to mean "in the entire program," but that's often not what it means in Cakewalk. If you turn your ProChannel off on one track, the button to do that is labeled "Global," but it only means that all ProChannel modules on that track will be bypassed, not in the entire project. With plug-in oversampling, "global" means that if you enable it in one specific plug-in, let's say elysia mpressor, all instances of elysia mpressor will be oversampled. The "2X" button better fits my usual idea of "global."

 

[Teegarden]

2 hours ago, Starship Krupa said:

If you take a look at the results of the tests I ran, I came to the conclusion that if you want to experiment with the possible benefits of oversampling, running at 88.2 incurred a significantly lower performance hit than enabling all the plug-ins. But the plug-in that caused the biggest hit (no pun intended) was a synth, not a processor. My Plugin Alliance elysia mpressor, alpha compressor, and Millenia NSEQ caused much less performance hits. AIR Hybrid, much less of a hit than A|A|S Player.

What I don't get is that you get better results with the whole project at 88.2 kHz than with just a few plugins at 2x oversampling. This is also in contrast to the Reference Guide, that seems to indicate that the opposite should happen (which is also more in line with what I would expect). Could it be that your particular hardware/DAW setup and project is diverting so much from an average project and DAW PC that you get other than expected results (can't think of anything specific, but it seems there must be something different. I've seen weird unexpected things on PCs, usually older, less performant ones, where a particular part of the hardware led to other than expected results)? 

 

2 hours ago, Starship Krupa said:

One of the things that led to confusion when I first started using Cakewalk was its use of the term "global." I've always understood that to mean "in the entire program," but that's often not what it means in Cakewalk. If you turn your ProChannel off on one track, the button to do that is labeled "Global," but it only means that all ProChannel modules on that track will be bypassed, not in the entire project. With plug-in oversampling, "global" means that if you enable it in one specific plug-in, let's say elysia mpressor, all instances of elysia mpressor will be oversampled. The "2X" button better fits my usual idea of "global."

Thanks! I've been breaking my head over Cakewalks interpretation of the word "global"... I think the Reference Guide could be much clearer on things like this.

Still the question remains if using the "global" 2x setting on top of a 96 kHz project still has some added benefit, or that at that point it just becomes wasting resources.

 

 

[Starship Krupa]

30 minutes ago, Teegarden said:

Could it be that your particular hardware/DAW setup and project is diverting so much from an average project and DAW PC that you get other than expected results

Sure it's possible. It's a decade-old hand-me-down Dell Latitude E6410. About a year ago, I upgraded it by replacing its original i5 with an i7. I'm using its onboard (not Realtek) audio with WASAPI. I figured that even if the clock speed were slower, twice the cores and more cache would help with DAW and NLE use.

The A|A|S Player engine sounds amazing, probably because its algorithms are constantly crunching a ton of analog modeling. Likely similar with Phoenix Stereo Reverb. This makes sense, the more modeling of "real" objects and spaces, the greater the load. I usually kick resource-hungry plug-ins to the curb, but both of these sound better to me than anything else of their kind.

I'll repeat the test on my main system and report back. I sort of recall that one of Noel's systems is (or was) an i7 3770 like mine, so it may yield results closer to the documentation.

For me, this stuff is interesting to mess with, but I think the true test is to throw on my best set of cans, upsample every plug-in for playback, and A-B test playback with the 2X button. Or  upsample them all for rendering and make two exports, one with and one without. If they sound different, even if it's a placebo effect, then the can of worms is open.?

[Starship Krupa]

Oh dear, I was afraid of this. Curses on you @LittleStudios! ?

I'll write more later, but preliminary results indicated an audible benefit to either enabling 2X oversampling of the plug-ins or rendering the project at 88.2. It's similar to the difference I can perceive between <256 MP3's and FLAC's. Individual elements sound more discrete and there's more apparent depth. Switching the 64bit engine on while rendering made no change that I could notice.

I did renders of the same piece with and without upsampling at 44.1, then with and without 64bit. All renders were to 24-bit WAV's. Repeated the renders again at 88.2. So the potential for highest quality would have been 88.2 with 2X upsampling using the 64-bit engine. More about how that turned out later.*

I enabled upsampling on both the synths and the fx, I imagine that I could narrow down which plug-ins make the biggest difference via trial and error, and I would start with A|A|S Player and Phoenix Stereo Reverb. The audible difference seems to be the same for both states, the version with the 2X upsampled plugs sounds pretty much the same as the one rendered at 88.2, at least in the wee hours of the morning through my speakers. I'll do more listening tests tomorrow.

*Weirdly, I did have one synth plug-in severely and repeatably misbehave when I enabled both 2X upsampling and rendered at 88.2. The bass arp track, which is A|A|S Player, starts emitting bad notes. The notes still sound, but one of them is off key when it cycles through the arp. Another synth also possibly sounded off rhythm with those settings, but it might have been that the bass arp was the culprit and was throwing me off. They sounded fine with all the other permutations.

Conclusions so far: try rendering the same project at 88.2 or 96 and compare it to itself rendered at 44.1 or 48, because I hear a difference, and I was carrying a healthy skepticism going in. I'm actually biased against  hearing a difference, because I would rather not have to concern myself with more settings. I have the suspicion that the improvement comes at a certain threshold and that 4X upsampling or rendering at 192 wouldn't make a further difference. It might, though, I haven't tried it yet. I can't draw any conclusions from the 2X/88.2 render because it's out of tune. Except it does suggest that my concerns about possible negative effects were well-founded.

Now I'm really interested to know what @Noel Borthwick makes of my impressions.

Edited October 5, 2021 by Starship Krupa

[Mark Morgon-Shaw]

All these high sample rate , high oversampling options are pretty moot in a real world scenario.

Most people can't out-mix 48k 24bit...and if they can most listeners can't hear it or lack the equipment to resolve such n'th degree differences.

 

[LittleStudios]

4 hours ago, Mark MoreThan-Shaw said:

All these high sample rate , high oversampling options are pretty moot in a real world scenario.

Most people can't out-mix 48k 24bit...and if they can most listeners can't hear it or lack the equipment to resolve such n'th degree differences.

I think your statement is far too broad.  It really depends on why you're using oversampling.  In the reproduction of a wave, as long as the sample rate is double the desired maximum frequency, the analog reproduction of the wave will be accurate.  For example a sample rate of 48KHz can accurately represent a wave at a frequency of 24KHz.

The difference is when you start processing the audio with non-linear effects (compressors, saturation, etc.).  These effects can cause aliasing.  When the added frequencies exceed nyquist, they fold back into the audible range.  In this case oversampling can make drastic improvements. 

I've taken 5 minutes and through together a quick video demonstrating how audible this is.  I use a tone generator.  Run the signal through a saturation plugin.  I then sweep through the frequencies.  The first run is the oversampling disabled.  You can clearly hear the "radio frequency scrolling" effect when the frequency sweep is in the upper end.  Sounds like the old timey computer sounds they used to use on the Enterprise (Star Trek).  I then run a sweep with oversampling enabled.  The sound of the "radio frequency scrolling" effect starts much higher in the frequency band and is much much quieter.  I perform the test one more time with oversampling first disabled and then enabled.

Here's a link to the video: https://youtu.be/9yS1CTu4kJc.

Now, that is a single tone sweeping through the frequencies.  Imagine cymbals, vocal air, crunchy guitars, synths all contributing to this effect.  You'll end up with a lot of unwanted noise folding back into the audible range.

Edited October 6, 2021 by LittleStudios
Further explanation

[LittleStudios]

I've been conducting more tests with the built-in plugin upsample feature.  Beware, this feature does not appear to address/report it's latency back to the DAW the way plugins with their own built-in oversampling do.

I started a new thread for this topic: https://discuss.cakewalk.com/index.php?/topic/34286-feedback-cakewalks-built-in-plugin-upsample-does-not-address-latency/.  I provide a brief description of the issue at hand.

In this thread you'll find a link to a video I've uploaded.  If you don't want to read the brief description and just want the video demonstration click this link: https://youtu.be/LgupFqtLDHc

[Teegarden]

3 hours ago, LittleStudios said:

I've taken 5 minutes and through together a quick video demonstrating how audible this is.  I use a tone generator.  Run the signal through a saturation plugin.  I then sweep through the frequencies.  The first run is the oversampling disabled.  You can clearly hear the "radio frequency scrolling" effect when the frequency sweep is in the upper end.  Sounds like the old timey computer sounds they used to use on the Enterprise (Star Trek).  I then run a sweep with oversampling enabled.  The sound of the "radio frequency scrolling" effect starts much higher in the frequency band and is much much quieter.  I perform the test one more time with oversampling first disabled and then enabled.

Here's a link to the video: https://youtu.be/9yS1CTu4kJc.

Incredible! Never thought it would such a huge difference... Even with suboptimal/cheap hardware you will be able to hear this kind of artefacts. I thought it would be audible, but not as pronounced as you've shown in your video.

Like you say, if you add up all the different tracks contributing to this, you end up with a lot of unwanted noise. I can't imagine the horrible noise that you create when using plugins that produce harmonics, saturation or any type of analog non-linearity (I guess there are quite a few plugins that do so; personally I like to use something like Black Box HG-2 on most of my submix busses and I guess at the same time I also use many other FX that add to the equation).

So, the conclusion is that it is absolutely worth using oversampling at a 48 kHz project.

Although, if your next project is an SF movie, you clearly might want to avoid oversampling?

You did not test higher project sampling rates. In your test it seems like just 2x oversampling at 24bit/48kHz is enough to even out the artefacts. So how much more oversampling is still worth the resources (88.2, 96, ...192 kHz with 2x, 4x,...oversampling). How do we know where to stop? (just guessing that not everybody wants to play hours with sinus waves in order to find out...)