32-bit floating point

OPTOBOT

OPTOBOT
i know its been covered hear before but i have done a search and haven't found the answer i was hoping for...

When boucing in cubase, to my knowledge when you bouce in 32-bit floating point then when you re-import that audio file it is exactly the same volume as before??

The problem is that this doesn't happen for me! im fed up of boucing stuff then having to remix it back into my tune!

If i have a tune at 24-bit and i bounce stuff in cubase at 24-bit it decreases in volume by about 7 dbs, this is the same for 32-bit and 16-bit..

Also when a tune is in 24-bit, is is good practice to bounce stuff and use cubase to export at 16-bit or use some one of the waves plug ins on the master mix to export to 16-bit?

Any help

Pete
 

BeatNik

DJohn Mustard Project
My guess is that you're mixing the track at a lower volume than 0 (as normal then)...
When you then export and then import the wave... it'll be imported at the same level as the original mix... but then reduced by whatever level below zero you're mixing (my guess is about -7 then)

My tip is... solo whatever you it is you want then bounce it down with the master on 0 - be careful so you don't have a limiter inserted on the master or anything like that... (make sure too that it doesn't clip - so having an entire mixdown bounced at 0 may not be the best idea)... this is the easiest way...

However! :Smile3:
If you do have a section that clips at 0.0 (or you don't want to change the master volume) then export it at the highest possible volume it can be... then increase the volume relatively for the new audio channel (i.e. minus 4 out - +4 in channel)...
Remember that cubase has a 32-bit floating point engine so it won't matter if an individual channel clips (the sound itself is not affected)...
as long as it's not clipping in the master-out :Grin:
 

BeatNik

DJohn Mustard Project
oh yeah... the best thing to do if a mix is 24-bit is to export it at 24-bit (or 32-bit - remember the vst's are all running on 32-bit) then dither it to 16-bit externally in soundforge or wavelab etc.... :Smile3:
 

ChrisCabbage

Forum Member
It's clipping in the master out that's been confusing me a bit recently. If I'm mixing down (exporting) at 32 bit, surely that gives me some headroom? Or is it that I'm effectively overloading the DACs?
 

BeatNik

DJohn Mustard Project
Hmm.. i'm not entirely sure about that either, but i recon that it's Steinberg trying to be clever :Smile3: (and emulating an analogue mixer). Maybe it's so that you can send the entire track straight into hardware without your signal becoming distorted...
Or maybe it is overloading the DAC's effectively drawing in an analogue clip?

Would like to know properly though? :Smile3:
 

BeatNik

DJohn Mustard Project
Purusha said:
That's kind-of what I'm thinking, but like I say - these things should be obvious...

agreed :Grin:.. would be good to just on with the creation process instead off faffing around with details like these, taking up your time...

But then again, these aspects are vital to creating a "perfect" mix (if ever one was to be made)...

-Don't make it less complicated, just make it more obvious :Wink3: -
 

OPTOBOT

OPTOBOT
the top audio file is the orriginal audio (the fader is at 0.0), the bottom track is the audio which has been exported (32-bit) then imported.

Its bloody quieter, aarrrgh :Sad:
 

BeatNik

DJohn Mustard Project
Are you sure it's not because you're exporting it with a lower volume on the master?

If you're not then I really don't know... from what I know that shouldn't happen
 

photonic ballast

Junior Members
I dont like that...Also happen with 24 bits? I think is Cubase Problem with your soundcard. I dont suffer that kind of things. If you can, take your project to a friend´s computer and render it there. Its not your fault :wacko:

About exporting to 32 bits and later in Soundforge reduce the SR...when using dither choose the triangular window and noise-shaping. Triangular dither is the most effective type.

Cheers.
 

BeatNik

DJohn Mustard Project
crap.. sorry yellowbrickroad didn't see that you'd written that the fader was on 0.0...

I really don't know then... try bouncing at different bitrates, and see what happens.

Sorry bout that :Sad: not fun when things aren't working as they should
 

OPTOBOT

OPTOBOT
xazzarr!!!

Sounds stupid but hadn't realised that the master fader was down a bit!!

Odd, cos i don't usually fiddle with the master fader, mmmm

Anywho yet another music tech problem that ended with me being stupid!!! doh

now, wheres the on button?
 

soliptic

whirling mathematician
some info on 32 bit that i posted elsewhere

i think some people are misunderstanding 32 bit slightly

the key factor is that its "32 bit floating point". whereas 16 and 24 are fixed point. so whilst 24 is 50% better than 16, 32 is not twice as good as 16.

think of a visual waveform (analog)

imagine you are a human A-D convertor. you are sampling this waveform manually. how? overlay it on a piece of graph paper. at every line verticaly down the page you trace up to the waveform and read out which horizontal line it intersects (or is closest to)

the X axis is time ( ----> ) so the density of our vertical lines equate to sample rate (khz). if we are sampling at 44.1 and our waveform is one second long then we will have 44,100 vertical lines to read a sample point at

the Y axis is amplitude , if we are working at 16 bit we have 65535 horizontal lines coming off this axis. (because binary 1111111111111111 = 65535). so we have 65535 possible different volumes we can store , which equates to something like 96dB of dynamic range (i forget what exactly)

at 24 bit we have binary 111111111111111111111111 or 16,777,215 different volumes, which i think is something like 160db of range.

but at 32 bit we move floating point which is basically like scientific mode on a calculator (1.045681 E12)... the first 24 bits stay as they are, but instead of representing an absolute volume, they represent a volume compared to some arbritary norm. the last 8 bits are used as a sort of 'scale factor' to determine how loud the sound actually is in the grand scheme of things. this gives us a dynamic range of something ridiculous like 1000db, and more importantly means that

(a) clipping is essentially impossible
(b) losing data by turning volume up/down or other dsp is essentially impossible

any decent audio software works at fully 32 bit floating point internally - mixer, routing, plugins, etc. (As for plugins themselves, I find practically everything i use is - a few a crappy fixed point, the only one i recall is autotune). if yours isnt - get one that is


but as for "my soundcard is only 24 so i cant hear or get any benefit from 32". basically if you think about the different methodology of floating point it is conceptualy impossible to get 32 bit data out of an A-D, or put it into a D-A. so recording and playback is always going to be dithered to fixed point, preferably 24 bit. but having a 32 bit engine in between is strongly recommended even if you only have 16 bit source material and final output, owing to (a) and (b) above. and bouncing audio tracks to 32 bit just ensures you are never leaving the floating point domain until you dither down your final mastered mixdown.

hope that makes some sense

this makes it a bit easier to understand

take this 24 bit sample

111011010000101110100011

make it half the volume - in binary halving is just a right shift

011101101000010111010001

and again

001110110100001011101000

now make it twice the volume - a left shift. but at the bottom, we dont know what the old number we lost was, so we just have to fill it with a 0.

011101101000010111010000

and again

111011010000101110100000

notice how we have now lost the last 2 bits of information for ever.

Now try it with32 bit

111011010000101110100011 * [ 00000100 ]

half the volume:

111011010000101110100011 *[ 00000010 ]

again:

111011010000101110100011 * [ 00000001 ]

double the volume:

111011010000101110100011 * [ 00000010 ]

again:

111011010000101110100011 * [ 00000100 ]

information lost: nothing.




couple of other points to pick up on

mixing in 32 bit *doesnt* mean you have headroom on the master, the master is effectively where it is dithered to a fixed point ready for your DACs. Never have the master clip, simple as that. On the other hand - clip your channels as much as you want, because they *are* provided with much headroom (nearly 1000db) by virtue of the 32 bit engine.

the only way a sample will lose volume when you export is if the master fader is down; if you are exporting through the master fader (which you do by standard, altho not necessarily) you will obviously have the gain applied as appropriate as well as any master fx you may have enabled. this is often dangerous territory as then when you re-import it goes through this twice
 

ChrisCabbage

Forum Member
OK - while we're here. If I put Brickwall Limiter on the master out, in theory I don't think I should see clipping, but I do. That's partly where I started to become confused...
 
Top