Comps at 96k?

[AudioAstronomer]

I was playing around at 96K today because I was trying out a new mic.. testing out on some drum tracks going for that pedro the lion "control" drum sound I love so much.

More or less the sound is in the compressor, ~10ms attack with 15-30ms release and ratio pretty well up on a mono room drum mic. 1176 nails the sound... saw's comps do too.

I compress like this on drums a lot and since I do 98% of my work at 44.1 Ive never said a word, it sounds great everytime and im a happy man!

But I digress a wee bit, today I was goin for the same thing at 96K and holy bejeebus it sounded amazing. I first thought it was maybe the mic... nope. Converters? nope. Went down to 44.1k and the "magic" was sucked out like a ... well i cant think of any clean analogies.

Im well aware of the AES and other papers on time domain process benefits from higher sample rates as Ive been researching for my own design, but this has gone way above my expectations. Im thinking perhaps the saw comp's algorithm maybe treating the sound to a different release shape (perhaps miscalculation?) or some sort of wonderful error?

I did some tests with noise, sine waves (simple and complex[tri,saw, etc..]) and at 96k the comp seemed to have a different release and attack shape drastically. Even a 96k sample comp'd and brought down to 44.1 was noticably different (and with music tests, much better sounding imo).

Any insight at all Bob? Because despite my complacency working at 44.1, this is totally worth the bandwidth and cpu usage to me. It really really sounds awesome and pretty much negates my desire for any other compressor.

[Kahalab]

Hey Man!

Thank you for the great tip about compressor settings for drums! I just applied them to some drums that i have been struggling with for ages and it did exactly what i have been (unsucessfully) trying to achieve! :) :)

[Bob L]

Give me an example of what to do to recreate what you are seeing and then perhaps I'll be able to see what's going on as far as calculations go.

Sinewave example that shows the difference in attack and release you spoke of.

Bob L

[AudioAstronomer]

Simply apply various compression settings to a Sine wave (I used 1K and 10K) at 2 different sample rates.

I tried it with white noise as well and it tended to exaggerate the difference.

Optimal test so far to notice differences that Ive seen is white noise peaking at -4dbFS and trying the compressor at -24 threshold, 10 ratio, 20ms release and 20ms attack. Changing attack/release seems to show the curve changing at higher sample rates. Try at 44.1 and 96k sample rates w/o conversion.


Just as a note, I really really like the way it sounds at 96k... itd be nice to hear some other peoples opinions on it as well. Awesome work at either sample rate Bob :)

[Dingo]

my view on 96k vs.44.1 k. 96k sounded immediately, obviously better and I have not looked back.

[dhise]

This is what I do: Get the best mix I can at 48k, then I set the upsampling to best quality and change the project to 96k and mix down. It always sounds better to me and it's not subtle either. The low end just tightens up and gets clearer. The explanation I guess is that non-linear processes like compression add harmonics and artifacts that go past nyquist that need to be filtered out properly. All the software compressors I use benefit from this and not just saw's.

-Doug

[Leadfoot]

This is what I do: Get the best mix I can at 48k, then I set the upsampling to best quality and change the project to 96k and mix down. It always sounds better to me and it's not subtle either. The low end just tightens up and gets clearer. The explanation I guess is that non-linear processes like compression add harmonics and artifacts that go past nyquist that need to be filtered out properly. All the software compressors I use benefit from this and not just saw's.

-Doug
That's not the same as if you recorded the audio tracks at 96 to begin with is it? It would seem that half the information would not be there, how does that work? If it does I'm gonna try it for sure.... curious.

Tony

[canipus]

Give me an example of what to do to recreate what you are seeing and then perhaps I'll be able to see what's going on as far as calculations go.

Sinewave example that shows the difference in attack and release you spoke of.

Bob L
just wondering, does the threshold value for the compressor work on peak samples or an integral over a time window or both? If there's any integration that works over a time window rather than a specific sample count you will get a different measurement value for the compression control depending on the sample rate. I presume if you're making any RMS calculations to capture the average energy that you're doing it over a set sample count right? In other words whatever time window the threshold is working in, the software compensates for the sample rate?

[AudioAstronomer]

CAnipis from what what Ive been seeing it seems like what you describe is the conclusion I'm coming to... it seems to be fairly predictably sample rate related.

All in all, I like it a lot... unless Bob can work some magic to get it working efficiently like this at 44.1k Im totally changing my viewpoint on sample rates.

[prosodio]

Mmmm, I´ll try a test... In the other hand, I think 1176 for tracking is great, Do you have one, robert?

Greetings :D

[AudioAstronomer]

Mmmm, I´ll try a test... In the other hand, I think 1176 for tracking is great, Do you have one, robert?

Greetings :D

yep :D more than one It's been my long journey to find something similiar in the soft world.. uad-1's is really close really really close, ive been working on a similiar comp to get that result I desire... but this is a very nice discovery.

[Jesse Skeens]

This is what I do: Get the best mix I can at 48k, then I set the upsampling to best quality and change the project to 96k and mix down. It always sounds better to me and it's not subtle either. The low end just tightens up and gets clearer. The explanation I guess is that non-linear processes like compression add harmonics and artifacts that go past nyquist that need to be filtered out properly. All the software compressors I use benefit from this and not just saw's.

-Doug

Bob is there any advantage to recording at 48k if you are going to do this? I thought the clean multiple from 48k to 96k was a myth now.

[Bob L]

I really don't have much to say about the high samplerates and the different methods to get there.

Everyone has their viewpoint on it, and I have found that many will constantly change that viewpoint from test to test.

I would say, do what sounds best to you at the time... when that changes, go with the flow.

For me... I am very happy with my current project results at straight 44.1k samplerates...

As far as the compression algorithm changing due to samplerate... I am pretty sure I have adjusted the samplecount for the transitions correctly to samplerate... but there are more samples at the higher rate, so each little level factor that gets applied during the attack and release are more subtle at the higher rates, because more of them are applied to get to the same end level factor... as to the proper ms length... in my tests here, the attack and release seemed to follow properly as far as overall average ms regardless of the rate.

Does it sound better? I'll leave that to the individual to decide. :)

Bob L

[AudioAstronomer]

Hum, the length seems to be proper, but the shape seems to change...

either way, this weekend after I get over this illness im going to do some double blinds to see if Im crazy or not. The difference doesnt seem to be subtle in the slightest, I was trying with the sonoris today and it also sounded different at 96... but not as drastically. Better? perhaps so.. perhaps not. But I sure like it :)

[canipus]

As far as the compression algorithm changing due to samplerate... I am pretty sure I have adjusted the samplecount for the transitions correctly to samplerate... but there are more samples at the higher rate, so each little level factor that gets applied during the attack and release are more subtle at the higher rates, because more of them are applied to get to the same end level factor...
Bob L

Yes but surely the algorithm is doing some integration and the individual differences wouldn't make that much difference because its the integral level that controls the gain reduction, (unless of course you're working on just peak detection). To be frank I wouldn't have thought the sample rate would affect the compression action that much tho' I'll accept the sound will be different by virtue of the fact the audio signal has twice the interpolated resolution. However, that will be a difference of detail and perceived clarity and would exist with or without the compressor being plugged in.