Login

Sample Rate Discussion (off of AITB Germanos EQ Thread)

If you need to write about anything else please do it here..

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby tumburu » Sat Jan 29, 2011 6:08 am

I'm not familiar with NAT so I don't know if it's possible. However, I assume that if it would be, then somebody would have already done that, don't you think?
User avatar
tumburu
Vip Member
Vip Member
 
Posts: 430
Joined: Tue Mar 30, 2010 11:00 pm

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby rrrobo » Sat Jan 29, 2011 7:23 am

You would think so! Maybe they have.
rrrobo
User Level V
User Level V
 
Posts: 52
Joined: Mon Jun 07, 2010 10:20 am

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby giancarlo » Sun Jan 30, 2011 2:03 am

rrrobo wrote:
I believe the solution is to have the option of buying 44 and/or 96 libraries as AITB intends to offer.


It is, but the real problem with this is having to capture the device twice - once natively at 44.1 and once again at 96. It would be very time consuming and not particularly inviting for the developers to have to do. I wonder if there is any advantage to converting the sample rate outside of Nebula with a high quality SRC like iZotope 64bit? I've asked this a couple of times before but never really got a definitive answer. Would you still get the artifacts mentioned? If not, developers could potentially sample once at 96 and then convert to 44.1 with no real loss or degradation (but of course it wouldn't be the perfect solution).

Cheers


I think r8brian could be better. Don't forget our algo is aimed to fast sample rate conversion (thousands of samples, sometimes even pretty long ones). Library developers could use other tools for a proper conversion. I never compared with r8brian really. For sure our approach is very good for music, but you know, impulses are a different beast. Sometimes they are really short (few samples!), so very difficoult to convert properly. Please note that if you start from 44100 and you are trying to get 96000, you don't have enough data in your original samples. For example you don't now at all what happens between 22050 and 48000. And there are AD DA filters, so 22050 is way theorical (even using the steepest filter possible)

Best thing for a developer would be sampling twice. You could sample @96000, than reduce it, but the result could be not so good. If you need 44100, the best approach is sampling @ 44100.
User avatar
giancarlo
Founder
Founder
 
Posts: 9195
Joined: Mon Sep 21, 2009 10:40 pm
Location: Italy

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby biomuse » Sun Jan 30, 2011 8:20 am

giancarlo wrote:Don't forget our algo is aimed to fast sample rate conversion (thousands of samples, sometimes even pretty long ones). Library developers could use other tools for a proper conversion. I never compared with r8brian really. For sure our approach is very good for music, but you know, impulses are a different beast. Sometimes they are really short (few samples!), so very difficoult to convert properly...

Best thing for a developer would be sampling twice. You could sample @96000, than reduce it, but the result could be not so good. If you need 44100, the best approach is sampling @ 44100.


As someone who works at 44, I'd really like to hear from Michael and Alex on this issue. I've been considering imminent purchases from both of them, but if the official word from the developer is that using the internal Nebula SRC means I'm basically not getting what I purchase, I need to rethink that and wait for some kind of resolution.

There's not much point in using Nebula to get some kind of erroneous approximation of the sound of a piece of gear. You can get that from algorithmic plugs.

44.1 is very commonly used. Given that, there should be versions of all products that work as intended at that sample rate.
biomuse
User Level X
User Level X
 
Posts: 119
Joined: Tue Dec 14, 2010 9:37 am

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby fuseburn » Sun Jan 30, 2011 10:07 am

Listen for yourself. You have to hear the artifacts and then decide if this is acceptable for you. It's been like this for years and many people have made great mixes with those "erroneously" resampled programs. If you don't hear a difference, be happy and don't waste another thought on it.

Sounds like a good solution, 44.1 and 96k. If you work at > 44.1, just upsample your material to 96k - powerful systems are cheap these days (seriously !).
User avatar
fuseburn
User Level X
User Level X
 
Posts: 133
Joined: Mon Apr 05, 2010 6:19 pm

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby biomuse » Sun Jan 30, 2011 10:32 am

fuseburn wrote:Listen for yourself. You have to hear the artifacts and then decide if this is acceptable for you. It's been like this for years and many people have made great mixes with those "erroneously" resampled programs. If you don't hear a difference, be happy and don't waste another thought on it.


This pretty much misses the point.

The reason I'm buying Nebula and its libraries in the first place is because I don't have steady (or any) access to the broad variety of hardware that is being sampled. I'll repeat: part of the value of the libraries to me is that there is some kind of direct, reliable correlation between the sound of the sampled hardware and its emulation in Nebula.

Giancarlo has now said that the SRC in Nebula is not optimal for minimizing artifacts, and he and others have confirmed that there are differences in sound that are not subtle when mixing sample rates. That tells me this is a real and substantial issue regarding the performance of the software. Saying "don't worry, be happy!" doesn't quite cut it. Again, if I wanted that, I'd just go back to using algorithmic plugs.

This issue is solvable in several possible ways, and so it would be better for the platform if it were solved. I would still like to hear from the major library developers on this.

fuseburn wrote:Sounds like a good solution, 44.1 and 96k. If you work at > 44.1, just upsample your material to 96k - powerful systems are cheap these days (seriously !).


I have a 2010 8-core mac pro, and for the kind of mixes I do (elaborate), 96k would still be an uncomfortable stretch in many cases. I'm not alone in that.
biomuse
User Level X
User Level X
 
Posts: 119
Joined: Tue Dec 14, 2010 9:37 am

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby giancarlo » Sun Jan 30, 2011 11:35 am

biomuse wrote:
giancarlo wrote:Don't forget our algo is aimed to fast sample rate conversion (thousands of samples, sometimes even pretty long ones). Library developers could use other tools for a proper conversion. I never compared with r8brian really. For sure our approach is very good for music, but you know, impulses are a different beast. Sometimes they are really short (few samples!), so very difficoult to convert properly...

Best thing for a developer would be sampling twice. You could sample @96000, than reduce it, but the result could be not so good. If you need 44100, the best approach is sampling @ 44100.


As someone who works at 44, I'd really like to hear from Michael and Alex on this issue. I've been considering imminent purchases from both of them, but if the official word from the developer is that using the internal Nebula SRC means I'm basically not getting what I purchase, I need to rethink that and wait for some kind of resolution.

There's not much point in using Nebula to get some kind of erroneous approximation of the sound of a piece of gear. You can get that from algorithmic plugs.

44.1 is very commonly used. Given that, there should be versions of all products that work as intended at that sample rate.



here we are speaking about perfection. We are comparting a downsampled 96K library with the same library sampled natively at 44,1K. They are degrees of perfection.

Guys, I think this topic is creating HUGE confusion. Sample rate conversion in nebula is terribly good, obviously you can't get perfection and at the same time speedness. It's still a compromise. A 96Khz library is very good even if you downsample it to 44100. It could be better if it was sampled directly at 44100. But maybe a library sampled at 96Khz and downsampled to 44,1 could be better than a native 44,1 library created by an other developer. Sample rate conversion is just a piece of the puzzle. For example HARDWARE TUNING BEFORE SAMPLING and other details (which hardware, jittering, adda filters, hidden issues in sampling gear).
Don't be too much anal
User avatar
giancarlo
Founder
Founder
 
Posts: 9195
Joined: Mon Sep 21, 2009 10:40 pm
Location: Italy

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby futur2 » Sun Jan 30, 2011 1:07 pm

i think there should be a better SCR alorithm implemented in nebula. even if the loading times are longer. or a choice between a fast/better and a slow conversion. i always thought nebula is always aiming for the best!?

the other thing is that there should be a simple way (nat or external) to permanently convert your 96khz libraries to 44.1/48. i know that some people tried this on their own and failed with lots of libraries. whatever the reason is? if not possible otherwise this permanent conversion should be provided by the library manufacturers.

giancarlo wrote:
biomuse wrote:
giancarlo wrote:Don't forget our algo is aimed to fast sample rate conversion (thousands of samples, sometimes even pretty long ones). Library developers could use other tools for a proper conversion. I never compared with r8brian really. For sure our approach is very good for music, but you know, impulses are a different beast. Sometimes they are really short (few samples!), so very difficoult to convert properly...

Best thing for a developer would be sampling twice. You could sample @96000, than reduce it, but the result could be not so good. If you need 44100, the best approach is sampling @ 44100.


As someone who works at 44, I'd really like to hear from Michael and Alex on this issue. I've been considering imminent purchases from both of them, but if the official word from the developer is that using the internal Nebula SRC means I'm basically not getting what I purchase, I need to rethink that and wait for some kind of resolution.

There's not much point in using Nebula to get some kind of erroneous approximation of the sound of a piece of gear. You can get that from algorithmic plugs.

44.1 is very commonly used. Given that, there should be versions of all products that work as intended at that sample rate.



here we are speaking about perfection. We are comparting a downsampled 96K library with the same library sampled natively at 44,1K. They are degrees of perfection.

Guys, I think this topic is creating HUGE confusion. Sample rate conversion in nebula is terribly good, obviously you can't get perfection and at the same time speedness. It's still a compromise. A 96Khz library is very good even if you downsample it to 44100. It could be better if it was sampled directly at 44100. But maybe a library sampled at 96Khz and downsampled to 44,1 could be better than a native 44,1 library created by an other developer. Sample rate conversion is just a piece of the puzzle. For example HARDWARE TUNING BEFORE SAMPLING and other details (which hardware, jittering, adda filters, hidden issues in sampling gear).
Don't be too much anal
futur2
Member
Member
 
Posts: 303
Joined: Sun Mar 28, 2010 8:16 pm

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby fuseburn » Sun Jan 30, 2011 2:56 pm

biomuse wrote:Giancarlo has now said that the SRC in Nebula is not optimal for minimizing artifacts, and he and others have confirmed that there are differences in sound that are not subtle when mixing sample rates. That tells me this is a real and substantial issue regarding the performance of the software.


All I'm trying to do is to convince everyone of trying to hear those artifacts with their own ears, and then make their personal judgement. You seem to be confused by what you're reading here. Please find out for yourself what amount of artifacts we're talking about here :D
Giancarlo summed it up perfectly: "Don't be too much anal"

biomuse wrote:Saying "don't worry, be happy!" doesn't quite cut it. Again, if I wanted that, I'd just go back to using algorithmic plugs.


...and - without any traces of irony - you should really know why exactly you're using nebula instead of the algorithmic counterparts which are often way more user-friendly, less convoluted and more intuitive (I know Giancarlo is working hard on that, no doubt) and you get a lot more mileage out of your 2010 8-core mac pro than with nebula. You obviously know, otherwise you wouldn't be here :)

biomuse wrote:I have a 2010 8-core mac pro, and for the kind of mixes I do (elaborate), 96k would still be an uncomfortable stretch in many cases. I'm not alone in that.


That's not true. If your mixes are big and complicated, you know why, so you know what you're doing, otherwise you'd keep it simple and stay away from - from that perspective - unjustified hazzle/CPU-load. If you know what you're doing, freezing is no deal breaker if you're using the right DAW.
We're all looking for the perfect solution, and at the moment it means upsampling your projects to 96k. And wait for AITB to release the announced 96k versions. It certainly won't become any more user-friendly with a better built-in SRC either - it'll rather take longer for the programs to be converted, and the result is bound to be worse than staying on the programs' native SR.
You know the tradeoff between light CPU and quality, so don't be under the illusion that it'll change anytime soon :lol:.
Same-ballpark machine here btw. and I do big nebula-loaded 96k-stuff.
User avatar
fuseburn
User Level X
User Level X
 
Posts: 133
Joined: Mon Apr 05, 2010 6:19 pm

Re: Sample Rate Discussion (off of AITB Germanos EQ Thread)

Postby tumburu » Sun Jan 30, 2011 3:57 pm

The only inconvenience to me when working with 96k libraries is the loading time. They do sound on a par with the ones sampled at 44, never noticed any kind of degradation.

What I believe is truly necessary is that the libraries could be converted offline/NAT, and saved at a different SR, at iZotope/Voxengo resampling quality. That would be IMO the most clever thing.

And no, computers are not powerful enough for mixing with Nebula at 96 (mac pro 8 core here too).
User avatar
tumburu
Vip Member
Vip Member
 
Posts: 430
Joined: Tue Mar 30, 2010 11:00 pm

PreviousNext

Return to Other things

Who is online

Users browsing this forum: No registered users and 5 guests