Upsamling option questions

[Jakub]

Hello.

I have some questions about the upsampling for plugins (x2) option in CbB. What exactly is the difference between Upsampling On Render and Upsampling on Playback . I suppose that vith VSTi that are generating audio tracks fro midi tracks it would mean that I will hear the effect of the upsampling only after freezing the midi track and generating audio track from it? But how does it go with plugins that are operating on already recorded/rendered audio tracks? Should I check Upsampling On Render or Upsampling Playback in such case.

Another question is if the upsampling can cause conflicts with my interface. According to manual my interface supports 44.1, 48, 96 and 196 sample rates. I usually work with 44.1 so the x2 upsampling will result in 88.2 khz sample rate which my audio interface does not support. Does it mean that i should not use that option at all in 44.1 projects and need to use 48 projects?

[msmcleod]

Up-sampling on Render is when freezing, bouncing or exporting audio, whereas up-sampling on Playback is just when playing the audio in real time.   Up-sampling will almost double your CPU usage, so you may want it only for rendering.

Your audio interface doesn't really come into it. The way it works is, that audio is read from the track, then up-sampled to double the rate as it's passed through the plugins.  It's then down-sampled back to your project sample rate before being sent out to your audio interface.

Up-sampling can help reduce additional artefacts introduced at audible frequencies, when a plugin tries to add harmonics above the nyquist frequency (which is half your project sampling frequency).  This is generally only applicable to plugins that generate saturation or distortion, and the artefacts although visible on an analyser, aren't (IMO) usually that audible.

The down-sampling at the end won't introduce any additional artefacts as this is a simple process.

It's worth mentioning though that:

-  Many modern plugins do internal oversampling, which can negate any benefit of up-sampling.
-  Not all plugins will work at an up-sampled frequency - some may even crash, especially your project is at 44.1K  (88.2K is not a common sample rate).
-  Many plugins will sound exactly the same after you up-sample them, but will take up twice the CPU usage.

So my advice would be, check each plugin first to see if it works, and if it makes an audible difference - if it doesn't, don't bother.

[Craig Anderton]

15 hours ago, msmcleod said:

Up-sampling can help reduce additional artefacts introduced at audible frequencies, when a plugin tries to add harmonics above the nyquist frequency (which is half your project sampling frequency).  This is generally only applicable to plugins that generate saturation or distortion, and the artefacts although visible on an analyser, aren't (IMO) usually that audible...[snip] So my advice would be, check each plugin first to see if it works, and if it makes an audible difference - if it doesn't, don't bother.

I agree with the bolded part - if you can't hear it, don't bother.  However, I've found a major difference with quite a few synthesizers that generate harmonically rich sounds. To avoid the CPU hit, I change Cakewalk's sample rate as high as it will go with the plug-in (even 192 kHz if possible), render the track as audio, then ratchet the sample rate back down to whatever I was using. When you compare a synth rendered at the higher sample rate to the same synth sound rendered at a lower sample rate, the difference can be quite noticeable.  

[Jacques Boileau]

3 hours ago, Craig Anderton said:

I agree with the bolded part - if you can't hear it, don't bother. 

Asking for a friend of a friend of a friend etc.: if your earing does not reach very high anymore, be it because of age and/or abuse, will you ear the difference and could you need oversampling even if you can't ear the difference yourself?

[azslow3]

5 hours ago, Craig Anderton said:

To avoid the CPU hit, I change Cakewalk's sample rate as high as it will go with the plug-in (even 192 kHz if possible)...

"A concurrent" has finally implemented up-sampling (in pre-release), as usual (for him) with max flexibility, so up-sampling individual plug-ins or whole chains, up to 768k.
Not that there was no other options (project sample rate can be changed on the fly there and it is not fixed to the audio interface rate), but for performance that is nice ?

Sorry for off-topic...

[Craig Anderton]

6 hours ago, azslow3 said:

"A concurrent" has finally implemented up-sampling (in pre-release), as usual (for him) with max flexibility, so up-sampling individual plug-ins or whole chains, up to 768k.
Not that there was no other options (project sample rate can be changed on the fly there and it is not fixed to the audio interface rate), but for performance that is nice ?

Sorry for off-topic...

Not off-topic! But by "192 kHz if possible," I was referring to the system as a whole. Some plug-ins that are oversampled won't respond to higher sample rates because they're already at their limit.  

[Craig Anderton]

9 hours ago, Jacques Boileau said:

Asking for a friend of a friend of a friend etc.: if your earing does not reach very high anymore, be it because of age and/or abuse, will you ear the difference and could you need oversampling even if you can't ear the difference yourself?

What you hear are artifacts in the audio range. For example, the aliasing could be 4 or 5 kHz. Unless you've been mixing front of house for a heavy metal band the past few years, you'll be able to hear that

However...to be fair, at seminars when I've demoed the difference between "pure" sounds and sounds with aliasing, some people prefer the grit of the aliasing. I've been mixing a song for a Ukraine rally in Washington, DC that has people in it from Public Enemy and the B-52s (among others), and the producer insisted on keeping the aliasing in a string part - even though I offered to replace it.

Ultimately it's about art, not science!

[Billy86]

@Craig Anderton To avoid the CPU hit, I change Cakewalk's sample rate as high as it will go with the plug-in (even 192 kHz if possible), render the track as audio, then ratchet the sample rate back down to whatever I was using. 
 

is this as simple as going into Preferences, making the change, rendering, and switching I back to 44.1, which is what I use… and, short of “rendering out” as a track export, by rendering do you mean freezing? Thanks much.

[Jacques Boileau]

@Craig Anderton Is there a simple recipe to create this distortion somehow? The point being, if I don't know how to recognize it, it makes oversampling more of a guessing game. Not a good thing. But yeah, I hear over 5k... err I mean my friend of course! ?

[Craig Anderton]

53 minutes ago, Billy86 said:

is this as simple as going into Preferences, making the change, rendering, and switching I back to 44.1, which is what I use… and, short of “rendering out” as a track export, by rendering do you mean freezing? Thanks much.

Yes, that's what works for me. But as a caveat, I have an interface that doesn't get weird when I change sample rates.

[Craig Anderton]

22 minutes ago, Jacques Boileau said:

@Craig Anderton Is there a simple recipe to create this distortion somehow? The point being, if I don't know how to recognize it, it makes oversampling more of a guessing game. Not a good thing. But yeah, I hear over 5k... err I mean my friend of course! ?

Assuming you have access to the z3ta, it has the option to change the internal sample rate. With some presets, the difference is quite dramatic. Check out this video to hear the difference between running it at 44.1, 96, and 192 kHz with the factory Clav patch. This demo is actually more about reproducing sound accurately than the sound of aliasing, but if you have an older sampled string library, play some of the high notes. The artifacts you hear are what aliasing sounds like. 

I think I need to do a long-form video about this for my YouTube channel, it's a question that comes up quite often.

[bitflipper]

In actual practice, aliasing is hard to hear. SPAN can show it to you, but if the aliased frequencies are more than about 50 dB below the main signal, you won't be able to pick them out. You may, however, experience a vague dissatisfaction with how a track or mix sounds but not be able to put your finger on why.

That's why it's a good idea to

• Understand what causes aliasing *,
• Know which types of processors are capable of causing aliasing** and
• Which ones don't***
• Visually check for it with a spectrum analyzer ****

* Aliasing happens when frequencies are introduced that are beyond the capability of your project sample rate to handle. Those frequencies can be generated by a digital oscillator or by harmonic distortion. I know of no other circumstance.

** Anything that clips the signal is a candidate for aliasing. Hard clippers are the worst, followed by compression with extremely fast (< ~8ms, but frequency-dependent) attack times and high compression ratios. Anything that generates tones (e.g. a synth) that can exceed the Nyquist frequency can cause aliasing. Note that most synths mitigate this internally and won't require oversampling, but there are exceptions.

*** Most effects don't alias. No sample library made at or below your project sample rate is capable of aliasing on its own. Gentle compression with low ratios and medium-to-long attack times won't, either. Transparent EQs don't cause aliasing, even at extreme settings. As noted above, most synths don't have an issue. Even if they do, it'll be on specific patches that have a lot of high-frequency content, such as the Clav example Craig suggested. 

**** Note that just because you can see aliasing with a spectrum analyzer doesn't mean you can hear it. I've read some ridiculous threads on Gearslutz where people  endlessly debated whether a plugin with aliasing at -160 dB is inferior to one whose aliased content is at -190 dB. It's angels dancing on the head of a pin.

[Jacques Boileau]

I love thread like these where I learn stuff that elude me before. Thanks guys!

[Craig Anderton]

11 hours ago, bitflipper said:

In actual practice, aliasing is hard to hear.

Except for when it's incredibly obvious   But, that depends entirely on what's generating the audio. I've never heard aliasing with sounds coming from the outside world into a computer, only sounds generated within the computer.  I really should do a deep dive on this on my youtube channel someday.

Edited April 11, 2022 by Craig Anderton

[Craig Anderton]

Just to follow up...I checked out a bunch of virtual instruments for aliasing, and as rule of thumb, the older it is, the more likely you'll hear issues. There were several instruments I couldn't upsample, because they were already oversampled internally and attempting to upsample them put their sample rate into the stratosphere.

So this is to confirm that yes, if you don't hear anything, don't worry about it. But it's definitely worth checking whether older plug-ins oversample internally or not. There is one other wrinkle: some people think that a synth's oversampling algorithm won't give as good a performance as offline rendering, because the synth has to do it in real time. I'm not sure if it makes an audible difference, but I mention it just in case it matters.

[Jacques Boileau]

@Craig Anderton I listened to your video of the clav patch above. Easy to hear and HUGE difference with the different sample rates! Quite an eye ear opener video. Thanks!