lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
What algorithms up-/down-/re-sampling in Sonar?
Been checking out different sample rates, advantages/disadvantages, quality of conversions etc. What I found on some sites are to use specific software like SoundForge or WaveLab to convert before importing into daws. I mean strongly recommended to do that. All major samplers I used have had different options for this. Dimension has a sinc option for rendering etc. And daws doing resampling usually have a handful of algos to choose from. So what is Sonar using: a) importing - a bit confused here since only bit depth is asked for b) exporting downsampling like 48->44 c) exporting upsampling like 48->96 Thanks. EDIT: Link to a useful article: http://karma-lab.wikidot....g-a-sampling-frequency
post edited by lfm - 2015/03/29 08:21:53
Cubase Pro 9 with SA2015 as backup - W7 i7 2.8GHz 16G GeForce GT 730 - RME HDSP 9632 + AI4S
|
bvideo
Max Output Level: -58 dBFS
- Total Posts : 1707
- Joined: 2006/09/02 22:20:02
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 11:06:36
(permalink)
Maybe another approach is to test the programs under discussion to see which ones do the best job of sample rate conversion. Here's one effort to do that, though they limit themselves to testing just 96K to 44.1K. No clue is given about what algorithms the programs use.
|
Anderton
Max Output Level: 0 dBFS
- Total Posts : 14070
- Joined: 2003/11/06 14:02:03
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 11:42:10
(permalink)
bvideo
Maybe another approach is to test the programs under discussion to see which ones do the best job of sample rate conversion. Here's one effort to do that, though they limit themselves to testing just 96K to 44.1K. No clue is given about what algorithms the programs use.
According to those tests, SONAR does extremely well compared to other DAWs. Given that Cakewalk uses iZotope algorithms for time-stretching, it wouldn't surprise me if they use iZotope's SRC algorithms as well. [Edit: Wrong guess, see post #11]
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 12:41:28
(permalink)
Thansk guys...
..but if it's actually the same algos as stretching...how about Artist- which of the qualities are there for import/export situations?
And some algos do a good job on certain music material - and less good on others.
That's why resampling algos are by choice, and a number of them to pick in the same app.
Reaper has got 10-12 or so, and various quality alternatives to run realtime or render situation.
One size does not fit everywhere.
|
Noel Borthwick [Cakewalk]
Cakewalk Staff
- Total Posts : 6475
- Joined: 2003/11/03 17:22:50
- Location: Boston, MA, USA
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 12:41:56
(permalink)
SONAR uses high quality windowed sinc interpolation for SRC conversions on both import and export.
|
bitflipper
01100010 01101001 01110100 01100110 01101100 01101
- Total Posts : 26036
- Joined: 2006/09/17 11:23:23
- Location: Everett, WA USA
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 12:56:36
(permalink)
Engineers often use the term "FM" - Frickin' Magic - for any blackbox process they don't understand but that Just Works so there's no need to dig deeper. SONAR automatically resamples when needed (e.g. rendering a 48KHz sample library to 44.1KHz or importing a 44.1KHz file into a 96KHz project) and does it exceptionally well with no user input needed.
Is it iZotope code? Probably not. iZotope's company description states that they license DSP technology to ProTools, Logic and SONAR, but they explicitly list "dithering, noise reduction, time stretching, and audio enhancement". No mention of resampling algorithms. The time-streching code is in iZotope_Radius_x64.dll, but I see nothing with an iZotope name regarding resampling. Furthermore, Noel has described SONAR's sample-rate conversion as "custom", suggesting it was developed in-house.
 All else is in doubt, so this is the truth I cling to. My Stuff
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 12:57:10
(permalink)
Many thanks, Noel. My concern has been since I noticed that default I had in Reaper made a 44k original reference track of a professional recording was resampled running the project in 48 - and it created peaks 1.6dB higher, actually overs then on a pro recording. So it's obvious it's not all transparent. And then thinking if importing 20 track from another daw - a tiny bit unwanted artfacts start adding up - and may be clearly noticable on a project. It's like using a tiny bit worse preamp on every track - and what you end up with is clearly noticable.
post edited by lfm - 2015/03/29 13:06:12
|
brundlefly
Max Output Level: 0 dBFS
- Total Posts : 14250
- Joined: 2007/09/14 14:57:59
- Location: Manitou Spgs, Colorado
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 13:08:47
(permalink)
lfm
a) importing - a bit confused here since only bit depth is asked for
SONAR let's you choose import bit depth because file bit depths can be mixed in the same project. But to avoid having to re-sample files on the fly, all files must use the same sample rate, and it's determined either by the sample rate of existing audio in the project or by the default rate for new projects you have set in preferences.
SONAR Platinum x64, 2x MOTU 2408/PCIe-424 (24-bit, 48kHz) Win10, I7-6700K @ 4.0GHz, 24GB DDR4, 2TB HDD, 32GB SSD Cache, GeForce GTX 750Ti, 2x 24" 16:10 IPS Monitors
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 13:31:55
(permalink)
brundlefly
lfm
a) importing - a bit confused here since only bit depth is asked for
SONAR let's you choose import bit depth because file bit depths can be mixed in the same project. But to avoid having to re-sample files on the fly, all files must use the same sample rate, and it's determined either by the sample rate of existing audio in the project or by the default rate for new projects you have set in preferences.
Thanks. Just assuming then, not knowing, that the very best algo is used for import, since not time or cpu critical. But since one algo might be better for certain frequency content - a choice would be nice. In help file it mentions a range of quality settings for Audiosnap - and maybe platinum has a selection box for import, or? Thought I might do an FR on this - but if in bigger version of Sonar then that would be kicking in open doors.
|
Anderton
Max Output Level: 0 dBFS
- Total Posts : 14070
- Joined: 2003/11/06 14:02:03
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 13:46:34
(permalink)
lfm
My concern has been since I noticed that default I had in Reaper made a 44k original reference track of a professional recording was resampled running the project in 48 - and it created peaks 1.6dB higher, actually overs then on a pro recording.
So it's obvious it's not all transparent.
If you look at the test results for Reaper referenced in post #2, you'll see that Reaper's algorithms aren't particularly good and certainly don't stack up to SONAR, Logic Pro X, etc. I don't know if they've improved since those tests were made, but if not, it's not surprising you wouldn't find them transparent. Really, the only reason to offer a choice of multiple algorithms is if there isn't one good one. That way you can choose the one that's "less bad."
|
Noel Borthwick [Cakewalk]
Cakewalk Staff
- Total Posts : 6475
- Joined: 2003/11/03 17:22:50
- Location: Boston, MA, USA
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 13:50:42
(permalink)
SONAR's resampling DSP was built in-house and has been time tested over many versions. As I said its very accurate and doesn't introduce any audible artifacts let alone gain changes. There's no need for "options". It just works transparently when resampling is needed.
|
drewfx1
Max Output Level: -9.5 dBFS
- Total Posts : 6585
- Joined: 2008/08/04 16:19:11
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 14:16:44
(permalink)
lfm
My concern has been since I noticed that default I had in Reaper made a 44k original reference track of a professional recording was resampled running the project in 48 - and it created peaks 1.6dB higher, actually overs then on a pro recording.
The 1.6dB "higher" peaks were almost certainly just intersample peaks that were already present in the 44.1kHz recording. It's just that the SRC moves where the samples are on the waveform. It's not at all unusual for an SRC to cause you to see peaks between +1dB up to about +3dB compared to the original, but these peaks were present in the original - they just happened to fall between the existing samples.
 In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
|
drewfx1
Max Output Level: -9.5 dBFS
- Total Posts : 6585
- Joined: 2008/08/04 16:19:11
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 14:34:05
(permalink)
Anderton
If you look at the test results for Reaper referenced in post #2, you'll see that Reaper's algorithms aren't particularly good and certainly don't stack up to SONAR, Logic Pro X, etc. I don't know if they've improved since those tests were made, but if not, it's not surprising you wouldn't find them transparent. Really, the only reason to offer a choice of multiple algorithms is if there isn't one good one. That way you can choose the one that's "less bad."
Despite the really ugly looking stopband ripple on the Reaper SRC's, except for the really pathetic (!!!) Reaper 2 IIRx2 SRC, I wouldn't expect them to be a problem except when there was very high level, very high frequency content present in a signal.
In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 14:46:27
(permalink)
Noel Borthwick [Cakewalk]
SONAR's resampling DSP was built in-house and has been time tested over many versions. As I said its very accurate and doesn't introduce any audible artifacts let alone gain changes. There's no need for "options". It just works transparently when resampling is needed.
Really good to hear. Sonar import was actually among those I never saw do ampltiude changes on importing cd reference material - as a reference talking to Cockos about it. I didn't know for sure if you just normalized something in the end - so could not tell if algo was better or not, looking at amplitude alone. Cockos team just claimed - this is to be expected resampling, peaks will be different. I remember on of the last updates that Vsampler did(10 years ago, known to Sonarites) was adding a range of new resampling algos. Article linked to in OP, also states there are differences and one should really convert before using the crappy stuff in daws. Where you really have to make compromises is where you do things in realtime, I guess. And you have a better/heavier one for rendering than realtime playback. Also what they do in Reaper as well as Vsampler.
|
drewfx1
Max Output Level: -9.5 dBFS
- Total Posts : 6585
- Joined: 2008/08/04 16:19:11
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 14:57:03
(permalink)
lfm
Article linked to in OP, also states there are differences and one should really convert before using the crappy stuff in daws.
I wouldn't put too much stock in that article as there are a number of technical errors in it.
In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 14:59:23
(permalink)
Anderton
Really, the only reason to offer a choice of multiple algorithms is if there isn't one good one. That way you can choose the one that's "less bad."
When was reality that simple - nothing I ever saw.  Most obvious is realtime stuff that may be too heavy on cpu. That pretty much what I discovered using Ozone 2 and 3 ten years ago - still sounding digital. Building math for good algos is what you can base a company on - look at Softube. - What we provide is so good that you don't need anything else ... that is a bit too much sales talk
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 15:21:30
(permalink)
drewfx1
The 1.6dB "higher" peaks were almost certainly just intersample peaks that were already present in the 44.1kHz recording. It's just that the SRC moves where the samples are on the waveform.
It's not at all unusual for an SRC to cause you to see peaks between +1dB up to about +3dB compared to the original, but these peaks were present in the original - they just happened to fall between the existing samples.
..and if you use another algo that intersample peak is not there - so which algo is doing it right? I chose a more heavy algo, same type, in Reaper and 1.6 went down to 0.6 dB - still claiming nothing to do with algos? Only source material? As a reference I did imports to Sonar with no such overs - but did not know at the time if Sonar possibly did some normalizing afterwords. Noel has clarified algo is not doing that. 1.6dB - that is 20% - you get intersample peaks that size? I'm not buying that explanation - not of that size. And if it's there it shows on analog signal later as well. I haven't tested this case at the time on analog side, but professional equipment would, I guess.
|
Anderton
Max Output Level: 0 dBFS
- Total Posts : 14070
- Joined: 2003/11/06 14:02:03
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 15:28:18
(permalink)
lfm
- What we provide is so good that you don't need anything else
... that is a bit too much sales talk
Well, think about it for a second. If what goes into the algorithm sounds exactly the same as what comes out, I don't know why you'd need anything else...but if it doesn't sound the same, then you DO need something else! This is also a good time to mention SSL's X-ISM free metering plug-in, which indicates if there's inter-sample distortion that conventional meters don't show.
|
drewfx1
Max Output Level: -9.5 dBFS
- Total Posts : 6585
- Joined: 2008/08/04 16:19:11
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 16:03:28
(permalink)
lfm
drewfx1
The 1.6dB "higher" peaks were almost certainly just intersample peaks that were already present in the 44.1kHz recording. It's just that the SRC moves where the samples are on the waveform.
It's not at all unusual for an SRC to cause you to see peaks between +1dB up to about +3dB compared to the original, but these peaks were present in the original - they just happened to fall between the existing samples.
..and if you use another algo that intersample peak is not there - so which algo is doing it right?
I chose a more heavy algo, same type, in Reaper and 1.6 went down to 0.6 dB - still claiming nothing to do with algos?
Only source material?
Different algorithms can have different amounts of high frequency attenuation, phase shift, aliasing, etc. You can't determine what causes a level change without examining exactly what changed. I'm assuming you're talking about peak levels, where differences would not be surprising. Changes in RMS would indicate the SRC was changing the level.
1.6dB - that is 20% - you get intersample peaks that size?
I'm not buying that explanation - not of that size.
You can easily get intersample peaks of that size or larger, depending on the audio itself, and, for maximum intersample peaks, over how long a period you measure.
And if it's there it shows on analog signal later as well.
I haven't tested this case at the time on analog side, but professional equipment would, I guess.
Yes. In theory, a DAC should allow for a few dB of headroom above 0dBFS to allow for intersample peaks. In practice that might or might not be the case.
In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 23:30:01
(permalink)
Anderton
lfm
- What we provide is so good that you don't need anything else
... that is a bit too much sales talk
Well, think about it for a second. If what goes into the algorithm sounds exactly the same as what comes out, I don't know why you'd need anything else...but if it doesn't sound the same, then you DO need something else!
This is also a good time to mention SSL's X-ISM free metering plug-in, which indicates if there's inter-sample distortion that conventional meters don't show.
I won't argue with "sounds exactly the same" - but maybe consider that's theory, not real life. And think about this: One track conversion - let's say moving a project into a daw that will convert things. When doing listening, I usually spend an hour back and forth to even conclude what result is, comparing. Differences are subtle. Let's assume one track is inaudible to hear differences. Import 20 tracks and listen to sum of these on master - then there might be even disturbing artifacts as a total. And make it 40-50 tracks and it's even more obvious. I read an article on preamps, why studios spend $3000 on a preamp when most of us feel that $1000 preamps sound as good as anything. But this tiny extra bit of clarity on each recorded track makes quite a difference when coming to the total mix of it all. So why do studio recordings usually sound just that little bit better than your own stuff - look at all parameters. Such cumulative things may be part of reason. I read loads of posts over creating "air", and do eq, saturation stuff and I don't know what - trying to compensate for less good recordings. Conversion stuff may be part of it. I recently learned about resampling in Superior Drummer, I had an instability issue in 2.4.1, that was fixed now in 2.4.2. It only occured if not having project sample rate to 44k, since it's recorded at 44k. So that raised my concern again about resampling issues - if to maybe go back to 44k. Have to do some tests on this before deciding. I do maybe half of tracks as recorded analog stuff and half ITB. I'm just trying to learn my tools, and what is the best way to use them - not to create problems showing up in final mix. I bought this album that both had cd and some 24/96 master on a companion dvd. - This will be fun to see if there really is a noticeable difference I thought. So I fixed up some strap in my external DAC to sense 96k and went on with testing this. a) first I found that dvd player is restricted by industry to downsample to 48k - that was a bummer. b) so I went on looking for computer based players that do 96k, and do asio etc so my RME card get the proper data. This just to find that the supplied 24/96 analogue master they supplied sounded crap and dull as anything. So why they do this is mystery to me - it's supposed to be a bonus, isn't it bringing something extra to you? So you live and learn, don't you.  Thanks for the SSL tip, really interesting.I guess if doing x8 upsampling(with a good algo) or something within a plugin, you will also get digital values for hidden intersample peaks in there. So having a metering plugin with that in Sonar, I would use it. It need only be used for a listen through before final render. I guess it's a cpu hog.
|
lfm
Max Output Level: -53 dBFS
- Total Posts : 2216
- Joined: 2005/01/24 05:35:33
- Location: Sweden
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/29 23:39:04
(permalink)
drewfx1
Yes. In theory, a DAC should allow for a few dB of headroom above 0dBFS to allow for intersample peaks. In practice that might or might not be the case.
Thanks for your input. I read an article by some that promote doing all mixing at -20dB peak levels. They have arguments like that - DAC seldom sound good at full 0dBFs. Noise floor is so low anyway - so you would not loose anything doing that - just benefits. Haven't really checked this, but there might be something to it.
|
Anderton
Max Output Level: 0 dBFS
- Total Posts : 14070
- Joined: 2003/11/06 14:02:03
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/30 10:37:14
(permalink)
lfm
Let's assume one track is inaudible to hear differences.
Import 20 tracks and listen to sum of these on master - then there might be even disturbing artifacts as a total. And make it 40-50 tracks and it's even more obvious.
I read an article on preamps, why studios spend $3000 on a preamp when most of us feel that $1000 preamps sound as good as anything.
But this tiny extra bit of clarity on each recorded track makes quite a difference when coming to the total mix of it all.
I agree with this thinking 100%. I first noticed it with hiss and acoustic projects. Preamp hiss would not be audible with individual tracks, but add together 24 of them, and you could hear the difference. If I applied noise reduction to the tracks, each track went from "inaudible hiss" to "even more inaudible hiss" but the final mix was like removing a layer of dust from a painting. This also happens with amp sims and layering guitars. The resonances that are "baked" into the sims become additive and the more guitars you add, the worse it sounds. I worked hard to avoid this effect in the sims I did for Cakewalk. Hopefully you can layer them without fear  Part of the reason for debates about SRC is that before the days of 64-bit calculations, there were audible differences among SRC algorithms. (In fact the 48 kHz sampling rate was chosen for DAT because the record industry felt conversion between 48 and 44.1 kHz was sufficiently difficult that it would discourage digital copying.) This is one reason why those who first started experimenting with higher sample rates chose 88.2 kHz because the conversion to 44.1 kHz was simpler. However these days, it is possible to do the math extremely precisely when converting from one sample rate to another. So saying "I won't argue with 'sounds exactly the same' - but maybe consider that's theory, not real life" has validity. However, thanks to improved calculation engines and algorithms, real life is becoming if not identical to theory, then really really really really close.
|
drewfx1
Max Output Level: -9.5 dBFS
- Total Posts : 6585
- Joined: 2008/08/04 16:19:11
- Status: offline
Re: What algorithms up-/down-/re-sampling in Sonar?
2015/03/30 13:20:06
(permalink)
AndertonI agree with this thinking 100%. I first noticed it with hiss and acoustic projects. Preamp hiss would not be audible with individual tracks, but add together 24 of them, and you could hear the difference. If I applied noise reduction to the tracks, each track went from "inaudible hiss" to "even more inaudible hiss" but the final mix was like removing a layer of dust from a painting.
You would hear the same difference just by turning up the volume of an individual track. Adding tracks together cannot reduce signal to noise ratio unless there is more cancellation between the actual signals on each track than the noise. In the real world, the reverse is arguably more likely.
This also happens with amp sims and layering guitars. The resonances that are "baked" into the sims become additive and the more guitars you add, the worse it sounds. I worked hard to avoid this effect in the sims I did for Cakewalk. Hopefully you can layer them without fear
If by "resonances" you are talking about frequency response, frequency response cannot change by adding things together. It can, however become much more apparent after adding tracks together because they cover a wider range of frequencies.
This is one reason why those who first started experimenting with higher sample rates chose 88.2 kHz because the conversion to 44.1 kHz was simpler. However these days, it is possible to do the math extremely precisely when converting from one sample rate to another.
Agreed, but it wasn't just precision - available CPU power to do the calculations was a huge issue compared to recent years.
So saying "I won't argue with 'sounds exactly the same' - but maybe consider that's theory, not real life" has validity. However, thanks to improved calculation engines and algorithms, real life is becoming if not identical to theory, then really really really really close.
In digital audio, anywhere where you're between the converters it's all just math. There can of course be bugs, HW problems and people can always mess up the theory or leave stuff out. But when applied to digital audio, well, let's just say those "theory vs. real life" arguments sound nice in theory.
In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
|