I am very interested in learning more about what the problems at higher sample rates are. I am toying with moving to 96, so I am quite eager to review pros and cons of doing so in terms of audio quality. Can you please share some opinions or links, etc (if you don't mind, of course)?

Worrall’s video for Fabfilter is interesting but his position on 96 kHz is completely arbitrary.

He shows 96 kHz having less aliasing and harmonic distortion than 48 kHz in an “contrived” (as Worrall puts it) example much like many are accusing Fred of doing here, but then says, what if we run the same plugin in 48 kHz with 8x oversampling, which is essentially pitting 384 kHz against 96 kHz , and then says look at the results.

Well what if we oversample by 4x for the 96kHz one then?

But, its easy to see that Fabfilter was promoting its HQ mode in the original Saturn in that video, which was only an 8x option and would probably have an interest in talking down 96k so users wouldn’t try using this 8x HQ mode with that sample rate, which would be a fairly disastrous 768 kHz resource imprint

It’s important to realize Worrall is often reading a script which might be promoting a company’s agenda.

I think you missed the point: As you can get better results at 48 KHz (thanks to OS) than at 96 KHz, there is no point to use 96 KHz. Sure you can also OS at 96 KHz but if the aliasing artefacts are already low enough at 48 KHz + OS, there is no point to go to 96 KHz.

Alistair

48kHz + OS aside, is mixing at 96kHz is only strictly beneficial for plugins that create harmonics? Are there benefits to using 96kHz when considering EQ, or is 48kHz 'enough' for any EQ moves (even boosts at 30kHz or even 50kHz)?

I'm asking because while I am finding myself in agreement with the theory behind 48kHz + OS for plugins that create harmonic content, I am wondering if there are other benefits to mixing at 96kHz for plugins that do not do OS.

This is a very educational and easy to understand video.

It’s a misleading point, as it is only true if only using oversampling plugins.

But in every single case, a plugin or virtual instrument without oversampling will have less aliasing at 96 kHz than at 48 kHz.

Chances are most people producing music will use instruments and plugins included in their DAW and other third party options which will not have any oversampling and will contribute massive amounts of aliasing.

Just as he mentioned in the video, bumping up to 96 kHz buys an extra octave of harmonics before reflection, which is massive as it improves the harmonics of every note which contain the harmonics gained in that octave.

So there simply isn’t a strong argument that one plugin with oversampling engaged in a 48 kHz project with non-oversampled components somehow obliterates the advantages of 96 kHz system-wide with every non-oversampled component of the project.

It’s just a poor argument.

If just mixing from stems and only using oversampled plugins and non-destructive DAW operations, then that’s a perfect situation for 48 kHz. But its a horrible suggestion outside of that.

Normal digital EQs do not create any harmonics.
But let me ask you this: if you don't hear the difference, why bother?

Because you want to be able to achieve that 10/10 sound?

Now that is not achieved by using a higher sample rate, or by a certain gear.
It is only achieved, if you can hear the difference.
If you don't hear the difference, its just impossible to make the necessary decisions to achieve the goal.

So in case you want to hear the difference, it would be beneficial to learn how analog gear sounds. You don't need a sound guru claiming his plugin was sounding like analog. You need to hear the difference yourself.

And when you can hear it, you will not ask anyone else, but you will make the decision on your own, because you can judge the cost vs benefit yourself.

Modelled after great HW - such a shame:




4x oversampling

Yes that's why Tchad Blake never ever uses this plugin.
NOT.

Who likes the Distressor?



Nobody.

@ 48k ?

Yes, as many plugins A) don't have internal oversampling at all or B) have poor oversampling implementation.

FF clearly spent some time on OS, hence it's being marketed. That doesn't mean it'll be the same for every plugin. Far from it.

Didn’t want to ignore you since you did respond to me.

Ideally I do prefer -18dB harmonics tests in PD since I run basically end to end gain-staged.

Fred is conducting an extreme experiment, but I am finding his exploration interesting.

It’s his own thread, so I’m not really seeing the harm.

There might be, for example it's easier to achieve more natural frequency and phase response at high frequencies (but OS is not the only way of achieving that, example). Please see this post for audio examples (PSP Infinistrip eqs get cramped at low samplerates):

and this for a screenshot:
Also don't get me wrong, as somebody already said, OS is not everything, I actually like PSP Infinistrip a lot, I just have to be conscious of its drawbacks and switch to other solutions when necessary (and obviously, I would love a HQ mode in there).

This is not true. Many plugins don't alias at all.

First the "massive amounts of aliasing" is hyperrbole and if the plugins don't handle the bandwidth increase when they should, they are not fit for purpose IMO. Don't use them.

It is the responsibility of the plugin developers to take care of any bandwidth increase. If they don't, the plugin is not well designed. Keep in mind that this extra octave is entirely inaudible. It is only needed to deal with the bandwidth increase due to the processing.

It is a perfectly good argument. Just avoid crappy plugins. If you don't know which plugins are crappy (which indicates you can't hear the problem which indicates it isn't much of a problem anywah), just avoid all those emulation plugins unless you know how to test them.

There are plenty of companies like DMG Audio, Cytomic, TDK and many others that do know how to do things right. Stick to them and avoid the major brands that invest as much in "photo realistic" GUI's, marketing and other such useless stuff as they do in the quality of the processing.

It is perfectly valid argument if you understand it.

It is the absolute best suggestion. If the developers can't even get aliasing right, what chance do they have of getting the rest right? Not much. Avoid crap plugins and you can stick to 44.1 or 48 KHz like the vast majority of professionals do.

Alistair

Dealing with bandwidth increase due to non-linear processing is fundamental to proper DSP implementation. If the developers don't even get that right... why use their plugins?

Also, in many cases, a two fold increase in bandwidth isn't going to be enough anyway so simply working at 96 KHz won't solve the problem. It is simply not a good solution to a problem that should be taken care of by the plugin developers.

The only solution is to avoid badly implemented plugins. That would also kill off the bad actors. That's good for everyone.

Alistair

Yes, those harmonics are completely audible. Check my audio examples out on this thread: Be careful bouncing Tracks made in Sampler

I’ve run far too many experiments at this point which counter that video’s conclusions, and I have the receipts.

If Sampler aliases so badly, why do you use it?

Alistair

I don’t. I made that thread when it was first released.

That said, every instrument except Sculpture in its high quality mode, and most of the plugins show similar behavior in Logic, and I showed a few examples of it in that thread.

But producing at 96 kHz and printing at 192 solves the problem for me when it comes to the instruments.

But that’s a problem that many other DAWs and plugins have, but its impossible to recognize that if a person stays in the 41.1/48 kHz bubble.

There are plenty of plugins which offer something interesting harmonically which won’t noticeably alias at higher project sample rates.

Seems odd to throw those opportunities away just to live in that low sample rate bubble.

If system resources are too serious of a problem, then making a workflow that incorporates making commitments to bounces to move forward becomes a perfect option.

By the way, did you listen to the audio examples and find the aliasing “inaudible”?

Great!



I do recognise the problem. That is why I avoid crappy tools.

I'm pretty sure I understand the topic significantly better than you. Here is some education for you. From 2011 but still relevant today:

Digital Audio and Sampling Rates


For some clients I work at higher sample rates but it is extremely rare. But guess what, all the tools that work great at 44.1 KHz also work great at higher rates. Best of all worlds!

I test the tools I buy and don't buy stuff that isn't well implemented. I've been doing it that way for two decades it works great.

The vast majority of professionals work at 44.1 or 48 KHz. Your bubble comments are not only rude but ignorant too.

Messing with your creative workflow and efficiency to cater for crappy tools is backwards IMO. Ditching the badly designed tools is a much better options. It solves this and so many other problems.

I really don't understand why anyone would want to bend over backwards to cater for sub-optimal tools... The tools should be there to facilitate your creativity and workflow, not hinder them!

Alistair

This thread has been running for over 10 years, do we really need another one? Some very smart people in this thread:
Lets do it: The Ultimate Plugin Analysis Thread

Come one dude... Who doesn't have a sine wave at 19kHz at 0dBFS? I know I do. All the damn time.

Thanks for the headsup.
OMG.
Their EQ "Equality" shows ZERO noisefloor. Down to -200 dB. Zero.
Boosting the hell - still nothing. World record?

Do they have anything that saturates, too?

But we do have much worse things: transients, sibilants, hi hats... Having HF as low as -20 and using saturation (tape-style?) can lead to pretty obvious aliases at mids and even low frequencies, if using non-oversampled plugins.

I do agree that using 0dBFS test tones is not so nice as to use -6 or even -20 tones at HF, but i might assure you that some (most) plugins can generate aliases at a very audible level even when used with pretty low level signals... and now imagine transients and percussion peaks, which are not as simple as sines, which leads to even more complex problems...

I do agree that “Ultimate plugins analysis thread” was more than enough to reveal the problem, but... seems like it is not enough for some “use your ears” apologists, who actually don’t use their own, just covering their “**** it” approach with sophistry.

Yeah, "nice" is no category for us engineers.
Is there any plugin that has a good aliasing-protection and my test shows it hasn't?

The relevant question for me as engineer is not, or if it is nice, or something no "authority" has ever demonstrated to the laymen and therefore they just reject the new method, because the mathematical and technical reasons why it works are not understood - the only interesting aspect of the applied method is, if it allows me to draw a correct conclusion.

The method I am using has allowed me to make dozens of correct judgements about the AA-ability and so far not a single false judgement about the AA-abilities has been named.


Btw: every critic could develop an even easier, quicker and at least equally reliable method to identify aliasing machines...

Oh cool, tell me do you have a web presence where I can listen to some of your work?

Also, I have to say, I personally don’t see how working at 96 kHz forces someone to “bend over backwards”, unless they don’t want to upgrade their computer as technology improves, which is more an odd personal choice to me than anything.

In case you didn’t understand, the audible change from the extra octave of harmonics gained from going up to 96 kHz in my examples, is from the folded back harmonics which also move up an octave, leaving the harmonics which were clouded by that aliasing clear. I hope that makes sense. Class dismissed.

Plugin doctor should have a pink noise option for harmonic measurements.

A sinewave is better, because it gives the most clear results as can be seen in my measurements and it also has the benefit that it shows the composition of the harmonic content.

You demonstrate in almost every post that you have no knowledge of signal theory. Why don't you just STFU?
Please leave my thread!