KVR Audio

@midnight · Post by **@midnight** » Sat Apr 28, 2012 9:06 am

How important is it to oversample if for example you are just using a clean compressor and doing like 2-4db gain reduction? Do you feel like you should oversample all compression or is it only necessary when doing deep compression with lots of gain reduction?

Thanks!

(trying to get an overall cleaner sound on my next batch of recordings, don't want to overdo it though at the expense of unnecessary wasted cpu and long rendering times etc... trying to find a balance here between a clean sound, and being able to get stuff done quickly (and at zero latency whenever possible))

mandolarian · Post by **mandolarian** » Sat Apr 28, 2012 10:13 am

I'd say, in a word, no. That's not much GR.

With such small reduction, maybe you don't need a compressor at all.

Too radical, ok, maybe you need 5-10 compressors (buy them all) in series doing .5dB GR each with 8X oversampling.

Maybe you just need a clean limiter - with infinite oversampling, of course. Don't worry about the latency - that's your great great grand kids kid's problem.

Shy · Post by **Shy** » Sat Apr 28, 2012 12:15 pm

@midnight wrote:How important is it to oversample if for example you are just using a clean compressor and doing like 2-4db gain reduction? Do you feel like you should oversample all compression or is it only necessary when doing deep compression with lots of gain reduction?

You don't have to feel anything else, just listen

. In DC8C for example, it has no benefit, just maybe very tiny benefit when doing "clean compression" with high threshold. In another compressor, it may affect more. There's a very "clean" sounding compressor that does extensive oversampling (not selectable), but you wouldn't like its 2122 samples long latency

.

meloco_go · Post by **meloco_go** » Sat Apr 28, 2012 2:21 pm

Shy wrote: There's a very "clean" sounding compressor that does extensive oversampling (not selectable), but you wouldn't like its 2122 samples long latency .

Which one is that?

Shy · Post by **Shy** » Sat Apr 28, 2012 10:38 pm

SplitComp

aciddose · Post by **aciddose** » Sat Apr 28, 2012 10:42 pm

oversampling doesn't even have much of an effect until you start to get into major hard-clipping territory.

there are going to be minor differences with high-frequency content but it just isn't that important in the general case.

try for example running a drum loop through a compressor without oversampling. run it at 44.1k, then try it at 192k. can you hear major differences?

i wouldn't expect you could hear any difference.

("general case" doesn't include high frequency synth leads designed to test this kind of issue.)

camsr · Post by **camsr** » Sat Apr 28, 2012 11:12 pm

The only time I use oversampling on the compressor is when the material has "hard-edge". It's almost unnecessary in 99% of compression.

osiris · Post by **osiris** » Sun Apr 29, 2012 9:27 am

I noticed JCompshaper has oversampling on it. What does it do? (Besides chew up CPU)

tony tony chopper · Post by **tony tony chopper** » Sun Apr 29, 2012 9:34 am

A compressor normally has a smooth enough envelope, so hardly any high freq in it. In theory 2x oversampling would be useful (I've seen higher oversampling in some compressors, never understood why), but in practice I never noticed any difference. On the contrary, when used as a limiter it's more a problem.

bootsie · Post by **bootsie** » Sun Apr 29, 2012 10:42 am

oversampling is not the answer in this case
well, actually it is an answer but a very weak one ...

aciddose · Post by **aciddose** » Sun Apr 29, 2012 11:06 am

if you're using your compressor to hard-compress a signal at -48db with very short attack/release times, oversampling is the answer.

http://soundcloud.com/aciddose/example-hardcompress

with just the kick in there it isn't an issue. you can see the harmonics are generated up to about 30th though. insert some hihats or other high-frequency components like conga or whatever and boom, aliasing madness.

oversampling by 4x to 16x will fix this.

bootsie · Post by **bootsie** » Sun Apr 29, 2012 12:07 pm

oversampling is not the answer in this case
well, actually it is an answer but a very weak one ...

aciddose · Post by **aciddose** » Sun Apr 29, 2012 12:26 pm

sure. what other answer would you have then? you're processing an input where the fractional information is not available. do you suggest estimating? the source of the issue is that you're applying a simple non-linear function. what do you suggest to apply when anti-aliasing f(x)=x(2-|x|) ?

oversampling is the most CPU-efficient solution, and it works.

another solution would be "don't do that", the problem is though that some people do.

@midnight · Post by **@midnight** » Sun Apr 29, 2012 12:30 pm

aciddose wrote:if you're using your compressor to hard-compress a signal at -48db with very short attack/release times, oversampling is the answer.

http://soundcloud.com/aciddose/example-hardcompress

with just the kick in there it isn't an issue. you can see the harmonics are generated up to about 30th though. insert some hihats or other high-frequency components like conga or whatever and boom, aliasing madness.

oversampling by 4x to 16x will fix this.

how much gain reduction there?

aciddose · Post by **aciddose** » Sun Apr 29, 2012 12:34 pm

-48db. attack = 0.05ms, release = 2.5ms, non-linear application.

the compressor is modeled after a circuit i designed and it works perfectly. any arguments about "wah wah, intermodulation-distortion, blah blah blah" are 100% horseshit.

the distortion and modulation are intentional and desirable. the aliasing is the undesirable component.

Oversampling compression - necessary for mild/clean compression?