Oversampling oddities

[Drew Stephenson]

At the last virtual SOSage meet we actually got into a brief discussion about music production (I know, right?) and Martin Walker (I think) linked to a video about the plugin oversampling available in Reaper now: https://www.youtube.com/watch?v=GjtEIYXrqa8

Being a curious sort I dragged up a nascent project that just had half-a-dozen or so tracks and ran a back to back test with all the plugins running at 44.1 and then oversampled to 88.2.
I was actually really quite surprised at how obvious the difference was, as if a gossamer veil had been lifted between me and the speakers.

So this arvo I pulled up a much more complex work in progress and went through a fairly laborious process of doing the same thing*.

Three things were apparent on this, firstly was a recognition that I really need to do a lot more work on this mix.
Secondly, and more relevant to the thread, was a repetition of that 'lifting a veil' effect.
Thirdly, and this is the oddity bit, the loudness of the oversampled mix went up by 0.3 LUFS. I don't know why this should be but I wonder if a fast acting compressor at low frequency might be letting more of an initial transient through?


Anyway, I loudness matched the two files and auditioned again and there is definitely a clear sound improvement in the oversampled file.

The downsides to this oversampling are two-fold (as far as I can see):
1) CPU usage went from 14-15% to about 25% (on a 65 track project with over 100 plugins).
2) Render time went from 1 minute to 5 minutes.

But that seems a fair price to pay for the sonic improvement from my perspective, so if you're a Reaper user, and you've got CPU capacity to spare, I'd seriously recommend having a look at the video above and experimenting.

For the cynical amongst you I have uploaded the two versions of the track to soundcloud. I doubt you'll hear the difference in the web-player but they're downloadable as 24bit wavs to load into your DAW. You'll probably need to drop the oversampled file by about 0.3 dB to match the loudness.
No oversampling: https://soundcloud.com/blinddrew/fat-ro ... MjaRMXN2Y9
2x oversampling: https://soundcloud.com/blinddrew/fat-ro ... zkNrs8N5YO

* there were a few things like M/S decoders and simple delays that I didn't oversample because, as I understand it, it wouldn't make a difference.

[Hugh Robjohns]

The very small change in LUFS value is almost certainly due to the slightly different responses of dynamics processing between standard and oversampling modes.

The 'removed veil' thing is a common response -- I hear it too.

But do you think any average listener would notice or complain without the direct A/B comparison? Genuine question of your opinion.

[RichardT]

This is my experience working at 96 kHz - without doubt things sound clearer.

[Drew Stephenson]

The very small change in LUFS value is almost certainly due to the slightly different responses of dynamics processing between standard and oversampling modes.

I assumed that there would be some effect on dynamic but I was surprised to find it that high (accepting that 0.3 LUFS is not exactly 'high' ).

The 'removed veil' thing is a common response -- I hear it too.
But do you think any average listener would notice or complain without the direct A/B comparison? Genuine question of your opinion.

Good question. On its own I'd say probably not* but I do think we're into 'aggregated marginal gains territory' (as British Cycling used to say).
This 'veil' is something that I still consistently notice when comparing my mixes with commercial ones. So if this is a little step towards narrowing that gap, at the small cost of a bit of extra CPU, then I think it's worth adding to the toolbox.
Plainly there's still a barrel of other stuff I need to keep working on, but if I can keep making these incremental improvements then in total they might sum to something that is noticeably better than what I was doing before.

I hope.

* I haven't tried listening to the files on a 'normal' set of headphones yet. I was slightly surprised that I could still hear the difference even via the soundcloud web player.

[Drew Stephenson]

This is my experience working at 96 kHz - without doubt things sound clearer.

Hmmm. It might be time to rethink my recording approach. But as I have about a dozen projects on the go at the moment I might leave that to the next album!

[James Perrett]

If you don't want to go through all your plug-ins and individually change all your plug-ins you could just set your project sample rate to be 96kHz. Reaper will automatically upsample all your sources and you can render to whatever sample rate you need to at the end.

I've been doing this since Reaper started using the R8brain conversion algorithm a few updates ago. I wouldn't say that I've noticed a massive difference doing this but every little helps.

[Drew Stephenson]

If you don't want to go through all your plug-ins and individually change all your plug-ins you could just set your project sample rate to be 96kHz. Reaper will automatically upsample all your sources and you can render to whatever sample rate you need to at the end.

I've been doing this since Reaper started using the R8brain conversion algorithm a few updates ago. I wouldn't say that I've noticed a massive difference doing this but every little helps.

This would be a much smarter way of doing things compared to what I have done today...

[James Perrett]

If you are lacking in processing power then it may make more sense to only oversample with certain plug-ins but it is far easier to set the whole project to oversample.

[Martin Walker]

At the last virtual SOSage meet we actually got into a brief discussion about music production (I know, right?) and Martin Walker (I think) linked to a video about the plugin oversampling available in Reaper now: https://www.youtube.com/watch?v=GjtEIYXrqa8

Being a curious sort I dragged up a nascent project that just had half-a-dozen or so tracks and ran a back to back test with all the plugins running at 44.1 and then oversampled to 88.2.
I was actually really quite surprised at how obvious the difference was, as if a gossamer veil had been lifted between me and the speakers.

Yes, 'twas me, and I've done a lot of Reaper tests since then, and heard some distinct improvements with some plug-ins.

Personally I'm going to carry on doing individual per plugin oversampling, as for me it's mainly the preamps/distortion plug-ins that benefit (Airwindows and Soundtoys' Decapitator in particular), while the TDR compressors I use tend to already have their own integral oversampling options.

Martin

[Tomás Mulcahy]

Yes I think it’s better to test each plug-in for aliasing. DDMF has an app to do it. It’s quicker than doing blind AB listening tests and it’s definitive. I’d consider using something other than saturation to achieve certain sounds. Because that’s the hardest effect to do with minimal aliasing.

By the way, the TDR plugins are incredibly good.

[Drew Stephenson]

Sorry, DDMF?

[muzines]

Sorry, DDMF?

https://ddmf.eu/plugindoctor/

Makers of Plugin Doctor...

[Drew Stephenson]

Ah, thank you. I thought it was an abbreviation rather than a company.

[Martin Walker]

Yes I think it’s better to test each plug-in for aliasing. DDMF has an app to do it. It’s quicker than doing blind AB listening tests and it’s definitive. I’d consider using something other than saturation to achieve certain sounds. Because that’s the hardest effect to do with minimal aliasing.

I've now done SO many aliasing tests at 44.1kHz sample rate that I've adopted a standard test tone: 15kHz sinewave at a level of -18dB (most common default input level for non-linear plug-ins).

1. I can't hear this 15kHz tone at my age

2. With a 44.1kHz sample rate and a suitable saturation plug-in, the 2nd harmonic of this test tone will be at 30kHz, and its folded back aliasing tone at 14.1kHz (again inaudible to my ears).

3. However, the 3rd (typically the strongest harmonic for symmetrically modified waveforms) will be at 45kHz, with aliasing tone at a very audible 900Hz.

So, as soon as I start to create presets for any saturator/preamp etc., I pop in my 15kHz test tone, and if I can hear any 900Hz tone the it's time to turn on oversampling. More extreme saturation settings will typically require greater oversampling multiples, but there's no point in piling on ever-greater oversampling if you can't hear any aliasing.

By the way, the TDR plugins are incredibly good.

Agreed - that's why I rely on their internal oversampling. I generally leave them on the 'precise' quality, but switch to 'insane' before rendering, as I can just hear the improvement.

Martin

[Tomás Mulcahy]

[Matt Houghton]

Worth noting that Acustica Audio specifically recommend *not* using external oversampling option for their plug-ins, and especially not going higher than 96kHz, which they say can cause problems. (Their plug-ins typically work very differently from most others).

[Martin Walker]

(Their plug-ins typically work very differently from most others).

You can say that again!

[RichardT]

Martin, I notice that reverbs sound clearer at 96kHz. Have you ever looked to see if they create aliases at 44kHz?

[Martin Walker]

Martin, I notice that reverbs sound clearer at 96kHz. Have you ever looked to see if they create aliases at 44kHz?

Great question Rich!

No, they don't create aliases per se (reverbs aren't essentially 'non-linear' devices that create harmonics). However, since a reverb is creating hundreds of spaced copies of the original sound, it stands to reason that doubling or quadrupling the computing power you give the reverb would improve the clarity and transient detail of reverb-ed sounds, as well as improving any filter and EQ responses.

I too have tried out Reaper per-FX oversampling on my reverbs and have definitely heard this extra clarity on my 44.1kHz when the reverbs are running at 96kHz. Some reverbs even have integral oversampling options (such as my favourite 2C-Audio Aether, which even offers real-time and offline options, so you can further improve your reverb sound at render time).

It's well worth experimenting and using your ears to see what sounds best to you, but I'm sticking with per-fx oversampling rather than running my entire projects at 96kHz (or greater).

Martin

[Spells]

I am always dubious of having different real-time VS render over-sampling options set on plugins because I prefer to hear exactly what it is that I'll end up with.

Aside from that, I'm noticing a trend for people working at 48k now... with no relation to TV / video. I have made the same switch as working at 96k, on massive productions brought in issues but also, I found that on SOME material, lower sample rates kind of suited the music more. 96k on a stripped down acoustic piece is a great option, but I found on testing with modern pop type stuff, the lower sample rates sounded more 'right' somehow.