Some questions about bit depth

Hello, I have some questions about bit depth/resolution in Cubase.
Please let me know if I understood correctly each one of the following points:

  1. Cubase 7.5 works internally always at 32 bit floating point.
  2. When mixing inside Cubase, it is impossible to clip the single tracks (because of 32 bit float processing); caution should only be used on the master output bus (when doing mixdown… and perhaps also while mixing to avoid damaging the speakers?).
  3. Setting a bit resolution (e.g. 24 bit) inside the “Project Setup” window, will only affect recorded audio; VST instruments and plug-ins will always work at 32 bit floating point (except if they don’t support it?).
  4. If I do an audio mixdown at 32 bit float, and the project is setup at a lower bit resolution (e.g. 24 bit) as in point 3, the exported mixdown will have an higher quality than a mixdown at 24 bit (because the VST instruments and plug-ins will be processed and exported at 32 bit float). And no dithering would be required.
  5. When doing a 24 bit mixdown (with the project setup at 24 bit), the Apogee dithering plug-in could be used, because Cubase is lowering the resolution from 32 bit float (internal) to 24 bit. I have read though that it is not important to dither 24 bit files… but perhaps it would be better to do it anyway, since there is the possibility and is quick to do?

I have the project bit resolution (in Project Setup window) set at 24 bit, which is the maximum for my MOTU audio interface.
I am thinking that it would be better to export audio mixdown at 32 bit float files, for doing mastering on this files later (with Cubase or other programs).
Do you agree?

What about exporting 24 bit files for doing mastering later? Is it OK to save them without dithering, then applying a dither after mastering? Or is one export/mixdown without dithering enough to corrupt the audio irreparably? I’m asking because previously I have always saved pre-masters at 24 bit. But it would be also interesting to know this in general, also for 16 bit files.
In any case I think that I should not apply dithering more than once, right?

Thanks.

1 Like

Hey, Thomas,

I’ll try, to the best of my knowledge:

  1. Yes

  2. nearly impossible, yes but you can clip the output anytime, since you’re interface is outputting 24-bit. A gain plugin as your first insert on main bus can prevent audible clipping.

  3. Yes!

  4. There won’t be any audible difference in the outputted file, whether 32-bit or 24-bit. No dither required.

  5. there’s no need for dither. My mastering guy is ambivalent about whether I send him 24-bit or 32-bit. Just try to stay away from converting unnecessarily.

To add to 4), you have to realize that not only VSTs and plugins will be processed at 32f, but any audio with any processing done will be in the 32f domain. And a fader setting of -0.01 to an otherwise untouched audio track IS processing - not to mention pan, EQ, volume automation and stuff like that.

Some additions:

  1. The channels can’t clip but plugins can. Search for ‘gain staging’ on the web, it’s interesting and can get you more transparent mixes.

  2. As long as your mixbus doesn’t clip, 24 and 32 bit float deliver the exact same result. Can be proofed in a null test. If your mixbus (main out) goes over 0 dbFS, the 24 bit file will be damaged/clipped/distorted while 32 bit float can handle it internally even though your interface’ converters are clipping.

I do my work to 99% in 44/24 but export mixes in 32 float just out of habbit :laughing: 24 bit would do (below 0 dbFS of course).

Yes.

Any physical output requires integer data, like 24 or 16 bit. This means that anywhere the audio goes to a physical output, or to a file format that does, it must not exceed full-scale (FS), like for:
a) External sends
b) Outputs going to headphones or speakers
c) Export Mixdowns directly from tracks, where the target format is integer.


Yes.


32bit floating IS actually 24bit accuracy, but with an 8bit exponent. So there is NO difference under FS. Over FS, summing may result in the loss of a bit or so for the smaller signals, which is no great loss.


The resolution is NOT changing (see above). 24bits is basically well below the noise floor of most IO devices, so dithering is of lesser value. For consistently high level 16bit, its use may be limited, as the quantisation noise will be significantly masked.


You can use 32bit for the project, just make sure that anything going to the outputs of the MOTU doesn’t go over FS.
Export your files for the most flexibility, but not exceeding the formats required for where you are mastering it. If external, get their requirements. Remember that 32 files are actually 24 bit accuracy, so 24 bit files can be converted to 32 bit floating by the masterers without loss.

Supply the files to your masterer at your project sample rate. As the mastering engineer, Bob Katz has recommended, run your projects at the higher sample rate and only convert up when receiving external files, and only convert down at the very last, which is what the masterer can do, or you, if you prefer.


Yes.
ONLY dither at the VERY end stage, and it MUST be after the output level control. In Cubase that means it MUST be in either slot 7 or 8 of the main output, and MUST be the last plugin.


Dithering is only required to smooth the effect of abruptly dropping off bits as part of a bit depth reduction, say 24 to 16 bit. Audio is not corrupted by not dithering. Dithering is actually corruption that gives the illusion of smoothness, which is why it is best left to the end.

If you were producing 24 bit audio for video, there is no need for dithering. Basically, don’t do processing, expecially complex stuff like dithering, unless you HAVE to!

Think of dithering as breaking off part of a wooden stick. The end is now rough, so you would smooth it. If however, the break didn’t cause any roughness that you could feel, it would be unnecessary to smooth it.


Yes. ONLY at the very end, AND ONLY if necessary.

iZoptope provide a free Dithering with Ozone pdf.

While nominally directed at users of their Ozone plugin, the pdf is predominately about dithering generally, and in quite a bit of depth, though it is not deeply technical.

On page 10, there is a picture which is only two bits deep, so it is very flat looking. Beside it, is a picture that has been dithered, and while it is still only two bits, it has more depth (if a bit grainy). That graininess is the price of better apparent depth. This is why it should only be done once, because adding noise to something noisy begins to make the noise more noisy – severely diminishing returns.

Thanks to everybody for the replies.
I have a doubt: if dithering from 32bit-float (internal) to 24bit is useless, why does the Apogee dithering plug-in have an option for 24bit dithering?

Completeness?

Or just in case someone comes up with an audio interface whose converters have less noise than 24bits can register.

Actually, it is a good question, that may have a good answer.

Any other suggestions, or smartass comments?

It’s an ongoing debate in the field. Many of the top mastering guys do say, “dither at any loss of bit-depth”
http://productionadvice.co.uk/when-to-dither/

For instance, Waves’ Linear Phase plugins run 64-bit math and offer dither from their output.

I’m just gonna have to test!