Dither thing

[OberHeim-Kenobi]

Hello, I have some doubt around dither these days. It's for sure everybody is using it almost every time the file bit depth is decreasing. Well, i usually don't use dither from a 32 bit FloatP. mixing enviroment (LOGIC PRO) to a bounced 24 bit (32bit FP to 24 Fixed P).
It is when i'm mixing and i bounce to disk the whole mix.

------Is anybody using DITHER in that stage?

PTHD have a Dither mixing option that pulled me to think in that direction.(48 bit FIXED P. to 24 bit FIXED P.)

I have searching for that issue but i didn't found it. Sorry if i'm repeating.

[DesertDawg]

Interesting- I don't recall seeing this particular thing (32 to 24 re: dither) being discussed, either. You're still getting truncation, though, if you're not using dither on those bounces/renders (unless you're going out analog, then back in again for your bounces, but I imagine you mean that you're doing it internally).

Have you considered trying flat dither at that stage to see if it gives you any significantly different sonics? That way you're also avoiding applying any noise-shaping twice.

I"m also curious about why you're doing this... do you mean you're recording original tracks at 32-bit float, and then converting them to 24-bit for mixing purposes? If so, why start at 32, only to immediately add another stage of (debatably unnecessary) conversion?

[OberHeim-Kenobi]

Originally Posted by DesertDawg

Interesting- I don't recall seeing this particular thing (32 to 24 re: dither) being discussed, either. You're still getting truncation, though, if you're not using dither on those bounces/renders (unless you're going out analog, then back in again for your bounces, but I imagine you mean that you're doing it internally).

Have you considered trying flat dither at that stage to see if it gives you any significantly different sonics? That way you're also avoiding applying any noise-shaping twice.

I"m also curious about why you're doing this... do you mean you're recording original tracks at 32-bit float, and then converting them to 24-bit for mixing purposes? If so, why start at 32, only to immediately add another stage of (debatably unnecessary) conversion?

I start at 24 bit files, it is the all converters output (except 16 bit on a pressed button in my Apogee). The thing is all 24 files are mixed in LOGIC PRO (on a 32 bit floating point mixer). At this point, since this mixer ends in a bounced filed at 24 bit. It is a decrease bit depth.

I'm doing some research on my DAW before years of automatic innerce thing.
I will comeback here with my conclutions.

Thanks for you reply.

[tINY]

Unless you have your levels WAY too low, dither from 32 to 24 bit isn't an issue - Unless you have something to play it back on that has noise in the -140 dB range...




-tINY

[Bob Olhsson]

The problem is that the distortion produced by not dithering is correlated to the signal. You are trading 10 dB of distortion that will add up another 6 dB. with each additional truncation for a few dB. of additional noise that will only increase 3 dB with each generation of dithered signal processing.

Distortion accumulates and at some point, as digital audio gets reprocessed, it starts sounding crunchy and harsh. Dithering helps push that point to well after the digital volume control on somebody's player.

Because the level is so low, dithering to 24 bits isn't worth going crazy to do but if it's only a matter of checking a box, it's kind of silly not to.

[tINY]

Good point.

I don't work with a huge number of plug-ins and sub mixes. But I bet a lot of people do...

It can take as few as 8 trips through truncation (maybe 48dB) before you might be able to hear it in a sparse mix in a quiet control room. So, you may as well check the box. It shouldn't take too much processing power anyway.




-tINY

[DesertDawg]

Originally Posted by tINY

Good point.

I don't work with a huge number of plug-ins and sub mixes. But I bet a lot of people do...

I think what Bob was referring to is that truncation produces it's own noise - apart from any plugins or anything like that.

[tINY]

But each time you go through a plug-in, you can be truncated again.

If there is only one truncation from mic to master, then you get maybe than 1LSB of noise due to it. For a 16 bit system - it's down out -96dB (for 24 bit, -144dB).

Every time you go through a gain stage or an EQ or compressor plug-in (or whatever) you raise it up by something less than 6dB. So, 10 processing steps in a 24bit box puts the correlated distortion at a maximum of about -84dB.

This is still below the noise floor of most mics....




-tINY

[rhythminmind]

I do allot of digital mastering. I believe you should always dither. BUT you should only use "Noise shaping" dither once. You don't want to have a build up of highend noise from layering noise shapes.
I use triangle dither when going from 32 to 24 if i'm still going to be working/editing the files. I then use PWR-3 for the final master.