Login

why does rendering or bouncing lose mojo and 3dness in logic

Tips & tricks, working results, technical support

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby domandroland » Mon Nov 14, 2011 6:36 pm

Yeah it does it in cubase too, i tried awhile back. Before anyone else recommends anything obvious, I have tried all file formats and bit depths, online and offline bounces and recording directly back in on most DAWS internally and DA-AD.
---
www.domandroland.com
New singles with Noisia and Amon Tobin out soon.
domandroland
User Level I
User Level I
 
Posts: 11
Joined: Sun Apr 11, 2010 10:37 am

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby giancarlo » Mon Nov 14, 2011 6:51 pm

it should render the same thing... really I don't know...
User avatar
giancarlo
Founder
Founder
 
Posts: 9138
Joined: Mon Sep 21, 2009 10:40 pm
Location: Italy

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby TranscendingMusic » Mon Nov 14, 2011 7:17 pm

About ywo years ago I looked into why sounds were different in Sonar regarding bouncing and such. I realized it's not that there is a difference in the audio quality but that certain things apply differently (like pan law) depending on how you pick the source. Now, this applied to Sonar so may be there is a 50/50 chance it's what is happening in your daw as well.
.
Amongst the tests I did, there was the bounce test. And what I found was that if starting with a mono source, in which the results can be rendered mono, make sure to choose "Track" as the routing source. If stereo, then select "entire mix". If the two are mixed up, meaning contrary to what the source track is, that's when I would get different sounding bounces versus live sound. So by following this, I ensure a 1:1 null. Now again I'm not sure exactly what the protocol is fr your DAW, but may be explaining this would clear something up for you going about to get the same outcome from choosing the right parameters in your daw's bouncing procedure. If not, and you double checked that you are not truncating or doing anything detrimental during the process, then sorry man, I'm at a lost there.
mixing | mastering
Win 10 x64 | Sonar Platinum x64 | 3930K(OC)
User avatar
TranscendingMusic
Expert
Expert
 
Posts: 1132
Joined: Sat Mar 27, 2010 7:01 am
Location: USA

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby biomuse » Mon Nov 14, 2011 8:22 pm

domandroland wrote:Yeah it does it in cubase too, i tried awhile back. Before anyone else recommends anything obvious, I have tried all file formats and bit depths, online and offline bounces and recording directly back in on most DAWS internally and DA-AD.


I found that subtle "flattening" only went away once I began to use 32 bit fp for disk storage of all intermediate processed or bounced tracks within a session. I.e.:

24 bit recorded starting files -->processing (EQ, comp, fx etc.)-->32 bit fp freezes/bounces to disk, all the way through the session --> dithered 24 bit format only at final stereo mixdown

As far as I know, writing 32 bit fp files to disk is only supported by DP (at least on Mac).

This way of working is effectively equivalent to performing no intermediate rendering for the entirety of the session. Were you to do a session which actually had no intermediate rendering at all, you would get the same result.

If you haven't yet tried that workflow, I'd be interested to hear what you think of it.
biomuse
User Level X
User Level X
 
Posts: 119
Joined: Tue Dec 14, 2010 9:37 am

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby mathias » Mon Nov 14, 2011 11:22 pm

i thought it is common knowledge to stay in 32bit float till the end of your working process in the digital domain (if you want highest quality and are not bound to certain formats, because of ...)

i assume working in a daw that works with 32bit float internally and throughout its processing-chains.

that said, i am not sure about the math or processing-differences between live-play and offlinebouncing.
it is executed in different speeds and there are for sure differences in the rounding processes, maybe temperatures of the processor, which causes different error-avoiding-processes to come into play.

a computer is absolutely not a perfect machine. it is just that we don't know about these processes on the deepest level and how they are achieved, to give us the impression, everything is perfect.
i am not a specialist in these matters, but i can imagine, that these things can change the way we perceive the audio.
so go by your ears and decide what sounds best to you.
even when you deceive yourself, you go away with a better feeling, which is always a benefit :D .

samplitude has since years an option to record the liveoutput of the mixer to a file and there were long debates about the fact that it sounds different to offlinebounces.
differences are so small for me, that i did not take that too serious. i always found, when the music really takes you, it is just not that important.
for bad music, well....... :roll:

so make good music and it will come through,
mathias
system 1: windows 8 32 bit - samplitude prox, tracktion6, reaper
system 2: mac osx yosemite - reaper(32+64bit), tracktion6(32+64bit)

both systems on: macbook pro (late 2009), core 2 duo 3,06 ghz, 4 gb ram, graphic: nvidia geforce 9600M GT 512 MB
mathias
Beta Tester
Beta Tester
 
Posts: 2106
Joined: Wed Mar 31, 2010 12:25 am

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby mathias » Mon Nov 14, 2011 11:32 pm

for everyone who has deeper interest in the matter, here is a statement of a samplitude-developer,said about six years ago, concerning calculations in daws and how they may differ:

-As programmer (and scientist) I tend to the deterministical point of view: just adding
two numbers should give the same in all circumstances, and hence the basic mixing
task should be the same for all 32 bit floating point DAW's - if you take the simple adding
process as a common method for mixing: we know there may be other methods here.

-I also know that due to the number representation in the computer, there just is not only
a limited precision but also a certain "random factor" in calculations with floating point
numbers. This is basicallly caused by the representation of the compiled code in the
system (initialisation of floating point registers with more than 32 bit processing,
usage of registers versus usage of computer memory,...) and the execution of the
code in the real system (e.g influence of other tasks, e.g. with multitasking).
Both is partially depending on the used compilers, but of course also from the
programmming code the application is written with.
A recent example could be a problem with the master gain in the first v8 beta
versions, where it changed the bit depth even when just initialized with exactly
0dB (= factor 1.0 in floating point).

-An obvious step after summing before going to the output is reducing the bit depth
again to 24 or 16 bit. This consists of at least 3 steps: clipping, scaling and dithering.
Every step can be implemented in various ways.
*Clipping can be done straight forward or more sophisticated, both with possible
hazards for the audio signal - but of course hopefully or at least commonly
only present to loud parts of the audio.
*Scaling can be done before or after summing (both is e.g. possible to be influenced by
the user in Samplitude with the input gain options or the master fader). Since scaling
also applies to the input, there's again a lot of things that can be done differently.
Even the discussions about panning law settings may be put here. Since integer based
engines require scaling before summing I can easily imagine different appproaches
in other DAW's.
*Dithering: nothing much to say here: there are more algorithms than we can think of
here. A major "ditherence" may be its application for intermediate steps when
necessary.

So here I see many possible reasons for different behaviour and even mis-behaviour.
And often I have seen or heard the results - not only in other DAWs, but partially also
in our (some still may remember a dithering problem in version 5.1x).

-An often uncommented problem seems the timeline representation of audio
material. There sure is a difference between sample based or PPQ based
approaches. The first gives perfect audio reproduction but possible problems
with loops and BPM values ;) while the latter more easily may sacrifice the first
in favour of the BPM.

So to sum it up: yes, the simple adding of numbers should basically give the same
results, but even there may be a bit of voodoo here - although it might be hard
to realize the differences. The more obvious reason for sound differences
are decisions about the processing at input and ouput stage, with a lot of
different ways to go - and I neither mentioned the connection with audio drivers
and other parts of the system.
system 1: windows 8 32 bit - samplitude prox, tracktion6, reaper
system 2: mac osx yosemite - reaper(32+64bit), tracktion6(32+64bit)

both systems on: macbook pro (late 2009), core 2 duo 3,06 ghz, 4 gb ram, graphic: nvidia geforce 9600M GT 512 MB
mathias
Beta Tester
Beta Tester
 
Posts: 2106
Joined: Wed Mar 31, 2010 12:25 am

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby domandroland » Tue Nov 15, 2011 3:17 am

Interesting read. It explains the dither coming from 32bit fp down to 24 i saw in span for one. I'll have another bash rendering at 32bit fp when my ears have recovered from the last couple of days sessions. Both Cubase and Reaper do it on mac. Always thought it was dumb that logic doesn't do 32bit. I hope it is this, so i can finally put this problem to bed.
---
www.domandroland.com
New singles with Noisia and Amon Tobin out soon.
domandroland
User Level I
User Level I
 
Posts: 11
Joined: Sun Apr 11, 2010 10:37 am

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby exrovite » Tue Nov 15, 2011 9:01 am

yes musicgreator,
I agree that the audio sound converter will ad some changes to the sound.

But that is if you have to go out of the interface and back in. With RME you can get an audio loop back in the interface which excludes the converters.
exrovite
User Level II
User Level II
 
Posts: 20
Joined: Sun Aug 15, 2010 11:53 pm

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby yr » Tue Nov 15, 2011 11:34 am

@op- I tested the bounce/playback difference this morning using the following procedure: created a 10 track mix in Reaper (32bit/44.1kHz), each track with a 2x saturation plugin (non-nebula to ensure accuracy). Inserted an Impulse before the beginning of the song to make sure I could align the tracks accurately. I then bounced the mix and later recorded the playback in Wavelab (rooted internally by the sound-card).

I re-imported the bounced/recorded tracks (32bit), aligned them and measured the difference after flipping the phase: -138dBFS (which is below the noise-floor of any sound-card).

I think there are many possible explanations why you are hearing a difference:
- your daw is buggy, bounced tracks contain errors.
- the software you've used for recording the playback is buggy.
- you've used plugins that have some random/semi-random behaviour
- you didn't align the tracks properly when trying to null.

etc

I think listening tests are tricky in this case, because they have to be carried out properly in a controlled environment. For instance, if you are the listener, I would argue that the one responsible for the playback shouldn't know the results either (double-blind). Even if you do find a difference (either through measurements or listening tests), you need to decide whether it is a significant/meaningful difference.
Reuven | post-production & sound-design | scenography |
website | nebula presets
User avatar
yr
Beta Tester
Beta Tester
 
Posts: 438
Joined: Mon Mar 29, 2010 9:04 am
Location: Amsterdam

Re: why does rendering or bouncing lose mojo and 3dness in logic

Postby biomuse » Tue Nov 15, 2011 4:44 pm

...And on that topic:

http://www.motunation.com/forum/viewtop ... 06#p407735

I'm as shocked as you will be. And pleased, it must be said.

I subsequently realized that although the "going away of the subtle flatness" corresponded in time to my switching to all 32bit renders - that became possible due to the big core rewrite of DP (v.6.0 if I recall) with its new SRC algos and a lot of other new math.

YMMV. Or it may not: you may be a closet Swede. ;)

On the 32 bit thing, though: Yr, nulls are tricky when it comes to psychoacoustically perceptible stereo phenomena. Paul Frindle (S*L G/E designer) has this whole thread on GS where he talks about a room of experts hearing stereo image differences that nulled to some similarly ridiculously low amplitude. In the mixes, that difference is not below the noise floor per se, and moreover it's a matter of the difference in signal perceived by your two ears, which relates to spatial localization, something we are evolutionarily super-primed to detect. That's all I got; beyond that, take it up with Frindle. :shock:
biomuse
User Level X
User Level X
 
Posts: 119
Joined: Tue Dec 14, 2010 9:37 am

PreviousNext

Return to Working with Nebula

Who is online

Users browsing this forum: igbro and 3 guests