32 bit floating point and dithering in LPX 10.4

[gurner]

Hi Guys,

I've been wondering about dither, and 32 bit floating point, in mastering.   I've been exporting in 32 bit floating point, which confusingly gives me 2 files, one labelled 24 bit.

So I import that file into the mastering session, it appears with a 24 bit label too!  (I have imported the file labelled simply .wav which info says is 32 bit)

Anyway I then apply normal dither when bouncing the master in 2444 - can anyone clarify whats happening here?

 

Edited March 31, 2023 by gurner

[des99]

A 32bit float file contains 24 bits of audio data, and 8 bits of the “scaling factor”, which helps move that audio data up and down the available range, so there is not much point exporting a mix as 32f when it’s below 0dBFS, you’re gaining nothing.

The reason DAWS use 32f internally is so that they don’t have to worry about clipping. On the mix bus, you’re ending up as fixed point data below 0dBFS anyway for your audio interface and digital audio files, so there is no need to go above 0dBFS and thus no need for the extreme dynamic range supported by 32f data.

Edited April 1, 2023 by des99

[gurner]

Thanks, des99 - so the best quality I can get for my pre-master in Logic is 24bit?

What about dithering, should I dither the pre-master or only the master?

 

Edited April 1, 2023 by gurner

[des99]

Dither is most needed when you are reducing the word length of audio. If you're exporting 16-bit wave files from Logic for use on CD, for example, that's an obvious case where you'd want to dither. 

In the case of exporting to 24-bit, you're not really reducing the word length, so you don't really need to dither. In your case, I'd skip dithering on export, do your work on the premaster, and then dither once when you export to your final reduced-wordlength files (eg, 16-bit for CD etc).

If you have Logic set to high precision summing mode (64-bit), you might want to dither when you export to 24-bit from Logic - I'd run some tests to see how necessary you think this is. With 64-bit, any quantisation noise is so far from being able to hear it, it likely doesn't matter, but ymmv.

[David Nahmani]

Here's a recent discussion on dithering that should interest you: 

[des99]

23 minutes ago, David Nahmani said:

Here's a recent discussion on dithering that should interest you: 

Good thread. I remember a least another one along similar lines too. 👍

[gurner]

All good - so one more important question, I believe maybe Yoad Nevo said - does anyone know to confirm - Waves analog noise - as from the SSL bus compressor - can substitute for dither in this situation ?

 

 

Edited April 1, 2023 by gurner

[des99]

I'm not sure why you'd want this, as it wouldn't be noise shaped - thus be far noticeably noisier than dither.

[gurner]

Hi des99 I believe the reason would be, that as we're using the Waves SSL comp anyway, which has an analog function, which adds a 'pleasant low level noise' from there,  was I to use that would Dither still be necessary?  I'm mastering a melodic tech house ep right now, and I believe this is being done in the genre by another respected engineer - up until now I've just dithered every bounce

 

Edited April 2, 2023 by gurner

[des99]

Your audible noise is likely way higher in level than any quantisation noise, but adding dither to help the quantisation noise is not going to make it noticeably noisier, so whether you want to is up to you…

[gurner]

Hi des99,  

Well I've done it without dither but with Waves 'Analog' engaged, for the 2444 file,

I'm wondering if I even need to dither the 16bit 'CD' file?

Anyway it sounds great to me

 

 

[gurner]

In the end I'm dithering the last bounce : POWR-1

Seems a safe bet - thanks for the info!

[des99]

You're welcome! I would always dither when going from 24 -> 16-bit, personally.

[gurner]

Yes, I would too, what confused me a bit was going to 24 bit - from the internal Logic 32 bit floating point, so I've decided not to dither the 'pre-master' (in my workflow) but only dither the final bounce, to both 24 and 16 bit 44.1

And I'm hoping this is the best practise in this case