32 Bit Float Mastering Question

[engmix]

I have to send a mix in for mastering. I'm now using PT 10 and my session is set for 32 Bits. I tracked my stereo mix back into pro tools from my analog chain. I'm exporting the stereo file for mastering and PT is giving me an option for 32 bit or 24 bit. Since Pro Tools has flagged the audio file as 32 bits, should i export the file as a 32 bits or 24 bits or doesn't it mater?

Thanks

[Adam Dempsey]

Personally I'd set it to export at 24 bit and not over-think it, as long as your peak level is still under 0dBFS (not clipping) at the 24 bit output. It's been a long time since I've received a 32 bit mix and it's not what I'd call a standard interchange format - it's more of an internal working format.

[haryy]

Unless you perform analog mastering why chop off a valid 32bit file?

[windmillsound]

I'm presuming that the OP is refering to 32-bit floating point vs 24-bit fixed point.

In which case it would not be chopping anything off - a 32-bit float is approximately the same as a 24-bit fixed as 32-bit float uses 24 bits for the mantissa (the value) and 8 bits for the exponent (the scale).

Floating-point math is useful in audio because 1) it cannot be overloaded and 2) you can offload the work to the maths co-processor on the CPU.

But you do this at the slight expense of requiring 8 bits for the exponent. However, nobody to my knowledge uses 32-bit fixed-point math in DAWs.

The short answer is it really doesn't matter whether you send 32-bit float or 24-bit fixed.

[wado1942]

If you're just sending the stereo mix you recorded from your analogue chain to be mastered, you shouldn't have to export anything. Just grab the mix out of the recorded files folder and send the engineer that.

[haryy]

Originally Posted by windmillsound

The short answer is it really doesn't matter whether you send 32-bit float or 24-bit fixed.

So, if it doesn't matter why every program works internally at 32bits (or 64)?

[Laarsø]

Originally Posted by haryy

So, if it doesn't matter why every program works internally at 32bits (or 64)?

Because it's easier. The PC is designed to work in 32 bit chunks. It's harder to write eq filters for 24 and 48 bit busses. This is explained in 1999 in the famous white paper by J. A. Moorer, PhD, "48 bit integer processink beats 32 bit floatink point for pro audio applications."

www.jamminpower.com/…bit%20Audio.pdf


"...In integer computation [read: fixed-point dsp], there is generally a tradeoff between roundoff error and dynamic range. One of the complications in integer computing is that this tradeoff must be considered at every stage of the process, thus placing additional burden on the designer. In floating-point, this tradeoff is made automatically, which can have unanticipated side effects as well..."


Cheersø,
Laarsø

[haryy]

From cubase 4 manual:"Bit Depth
(Audio Engine
Output section)

Allows you to select 8, 16, 24 bit or 32 bit (float) files.
If the file is an “intermediate mixdown” that you plan to
re-import and continue working on in Cubase, we recommend
that you select the 32 bit (float) option.
32 bit (float) is a very high resolution (the same resolution
as used internally for audio processing in Cubase),
and the audio files will be twice the size of 16 bit files."

This seems to imply that it does matter.

[Rob Murray]

Originally Posted by haryy

From cubase 4 manual:"Bit Depth
(Audio Engine
Output section)

Allows you to select 8, 16, 24 bit or 32 bit (float) files.
If the file is an “intermediate mixdown” that you plan to
re-import and continue working on in Cubase, we recommend
that you select the 32 bit (float) option.
32 bit (float) is a very high resolution (the same resolution
as used internally for audio processing in Cubase),
and the audio files will be twice the size of 16 bit files."

This seems to imply that it does matter.

Twice the size of 16bit but not twice the quality. 32 bit float is still only 24 bits of information with an 8 bit multiplier kind of like a window telling you where those 24 bits are. This matters for clipping, and this matters to have a longer bitdepth for processing purposes, but after being mixed down, you're not really losing information with 24 bit fixed vs. 32 float. Also Cubase is saying this as a note since many people who don't know the difference between 16, 24, or 32float might just pick 16 since that's where the project will end up not realizing the advantages of working at 32bit float-point processing.

I would strongly suggest 24bit for sending to the mastering house as some systems still don't handle 32bit files whereas I don't know of any program, certainly not that would be in a mastering house that can't handle a 24bit wav file. Much like I would always send a stereo file rather than a multi-mono. It's more reliable, compatible, and less likely to run into problems, human created or other down the line.

[dcollins]

Originally Posted by Rob Murray

I would strongly suggest 24bit for sending to the mastering house as some systems still don't handle 32bit files whereas I don't know of any program, certainly not that would be in a mastering house that can't handle a 24bit wav file. Much like I would always send a stereo file rather than a multi-mono. It's more reliable, compatible, and less likely to run into problems, human created or other down the line.

While it's unlikely to be a quality question, I think a mastering house should be able to accept any format that is likely to be presented. I would recommend Wavelab for playback in this case as it seamlessly handles 32 bit.

It also allows you to have any combination of formats open at the same time and jump between them, which can be useful if you have, say, an MP3 mock-up of the sequence to refer to.


DC

[Rob Murray]

DC - I totally agree, and I'm fine on my end for playback. Just figuring, there's still a few guys out there on older systems, who knows if they still might prefer Sonic, PT, etc. for playback and just in case it seems the easier option when exporting rather than having any flickering chance of an issue, or slowdown down the line.

That said... 32 bit files having been floating around a bit (puns intended) and at this point, yes, it shouldn't really cause any problems whichever you use.

[texasguitar903]

So should we or should we not record, mix, and mixdown to 32 bit float? I just maintain 44.1 16 bit for everything....isn't that where it ends up on a CD?

[dcollins]

Originally Posted by texasguitar903

So should we or should we not record, mix, and mixdown to 32 bit float? I just maintain 44.1 16 bit for everything....isn't that where it ends up on a CD?

No. It's desirable to always record and mix and more than 16 bits. Maybe even >44.1k as well.

The only time you should go back to 16 bits is for a CD.


DC

[texasguitar903]

So I should take all tracks in say, 32 bit float 88.2k. And export into wave files with the same quality. Unless someone wants them different for mastering or mp3.s for crappy internet distribution.

[haryy]

Originally Posted by Rob Murray

Twice the size of 16bit but not twice the quality.

If we accept that it's at least 1% better, then there's no point arguing about it. Providing that any further processing will be ITB.

[wado1942]

Originally Posted by texasguitar903

So I should take all tracks in say, 32 bit float 88.2k. And export into wave files with the same quality. Unless someone wants them different for mastering or mp3.s for crappy internet distribution.

Your converters are 24-bit, so the raw files should be saved as such or you'll just be throwing away disk space. The architecture of the DAW is higher (32-bit float) so it's advantageous to do all your subsequent generations at this resolution. It's a good idea, however, to send your mastering engineer 24-bit flat dithered files because there's many different types of floating point and while he can probably open it, it's safer to send a sure shot... Of course, you can always ask what he'd prefer first.

[Wave theory]

If I read the OP correctly, the mix has gone out, and returned via an analog loop.

Assuming there was no processing after the signal came back from the insert, then the signal will be 24 bit anyway. Unless the OP has a 32 bit A-D.

[Lagerfeldt]

Logic Pro Help • View topic - Increasing Overall Mix Level Before Bounce

[diamondjim]

Originally Posted by Laarsø

Because it's easier. The PC is designed to work in 32 bit chunks. It's harder to write eq filters for 24 and 48 bit busses. This is explained in 1999 in the famous white paper by J. A. Moorer, PhD, "48 bit integer processink beats 32 bit floatink point for pro audio applications."

www.jamminpower.com/…bit%20Audio.pdf


"...In integer computation [read: fixed-point dsp], there is generally a tradeoff between roundoff error and dynamic range. One of the complications in integer computing is that this tradeoff must be considered at every stage of the process, thus placing additional burden on the designer. In floating-point, this tradeoff is made automatically, which can have unanticipated side effects as well..."

Cheersø,
Laarsø

Thanks for that info, Laarsø, intesting person and web site. BTW the link above is incorrect, it is actually
http://www.jamminpower.com/PDF/48-bit%20Audio.pdf

[Gazz99]

Originally Posted by Wave theory

If I read the OP correctly, the mix has gone out, and returned via an analog loop.

Assuming there was no processing after the signal came back from the insert, then the signal will be 24 bit anyway. Unless the OP has a 32 bit A-D.

+1

your final quality is affected by every point in the chain... so in this situation, if there is no processing in Pro Tools after the analog chain, then you would not achieve anything (other than using up HD space) by exporting to 32-bit (and I'm guessing you would have your sound card set up to record the audio from the analog chain at its maximum - most likely 24bit)

[MOBMastering]

I may usually accept any formats. 32bit floating may have some problems between some DAW. I'm using Samplitude and in the past, I had bad experiences with engineers who had brought 32bit floating files rendered by Pyramix. The best way to be sure having no problem between DAW is to bring 24bit files.