Apr 9, 2012
Dither is one of the most widely misunderstood elements of the digital audio chain.
People don’t know what it is, they don’t know when to use it, and they don’t know how.
This post aims to fix that, and make it really simple. But before we get started though, let’s quickly answer a few of the questions.
What is dither ?
Dither is low-level noise added to digital audio when reducing the bit-depth to prevent quantisation distortion.
Quantisation what ?
Imagine a really quiet audio signal, just poking its head above the quietest signal your digital recorder can resolve. It’s so quiet, it’s only using the last available bit.
Without dither, all the encoder can do is choose to have that bit switched either on, or off. So instead of an accurate reproduction of the original signal, you get a gritty, grainy, gated fizz instead – you’ve added quantisation distortion.
To hear this for yourself, I made a video demonstrating it – click here.
And dither helps… how ?
In the simplest case, you just randomly switch that last bit on and off. The result is white noise – hiss – simple dither.
Now add in your very quiet signal again, and instead of having to constantly decide whether to switch that last bit on and off for the incoming signal, all the encoder does is add it onto the random dither value, and encode that instead.
The result is a better representation of the very quiet signal. It’s bathed in dither hiss, but it’s no longer gritty, grainy or gated – it’s just a very quiet sound covered in hiss.
That’s what dither does. The gritty grainy gated fizz is quantisation distortion – and you can remove it by adding dither.
(Actually strictly speaking, you’re not really removing it, you’re just de-correlating it from the input signal – but I said I was going to keep this simple…)
In fact, dither actually allows the system to resolve sounds below the theoretical noise-floor, unlikely as that might sound.
Can we really hear the difference ?
All this stuff is happening to only the very quietest elements of the mix, though – so surely it’s too quiet for us to hear ?
The results can be plainly audible at normal listening levels on many musical signals. Like mp3 artefacts they’re subtle, until you notice them – but after that they’re not subtle at all.
Quantisation distortion makes things sounds brittle, gritty, and collapses stereo imaging. In comparison the correctly dithered versions sound fuller, warmer, smoother and more “analogue”.
So when do we do it, again ?
It’s necessary whenever you reduce a digital signal to a lower bit-depth – for example from 24 down to 16-bit – but also, whenever you process a sound using fixed-point arithmetic.
Let me say that again.
Whenever you process a sound using fixed-point arithmetic.
So if you only ever work in one audio application, which works exclusively with 32-bit floating-point DSP, and always saves it that way, you’re OK. Oh, and all your plugins need to do it the same way, too. And they need to do it right. And they need to do it in a way that’s 100% compatible with your DAW. And not have any bugs.
Now that’s not an impossible scenario, by any means – but the question is, how do you know everything is working right ? If any plugin isn’t working as intended, your audio could get messed up without you realising it at any point. If you use that plugin on multiple tracks, it could be happening multiple times in a mix.
The good news is, there are several things you can do to avoid this problem
1. Work at 24-bit as a minimum
Quantisation distortion at 24-bit will be much less audible. Notice I’m not saying completely inaudible, though. Depending on the material, sometimes you can hear it even at 24-bit. Especially if things get processed over and over again.
2. Always use dither
This is a catch-all. If you always use dither when bouncing or saving fixed-point files, you won’t get quantisation distortion. (Well, you might, with a really badly-written plugin – but hopefully that’s a very rare scenario.)
Hang on !
There are people who’ll tell you this is a bad idea too, though.
Because remember, every time you dither, you’re adding a little extra noise to the signal.
And (again, depending on the material) sometimes that extra noise can change the sound slightly, too. Especially when you take into consideration all the different “flavours” of dither – almost every manufacturer has its own variety – Waves, POW-R, Apogee…
Relax – its just noise
All this debate is a red herring, though. Just stick with simple “triangular” dither, and you’ll be fine. After all, it’s just noise – and very quiet noise, at that.
All your favourite recordings from the 50s to the 90s were bathed in hiss – and they still sounded great. And that hiss was at a much higher level than any dither. Yes, dither may have a TINY effect on the sound – but you’re going to make much bigger changes with almost every other process you can apply, at any stage. Stop agonising.
When it comes to the final mastering stage, reducing down to 16 bits for CD burning, by all means pick your favourite fancy variety of dither – but before then, just switch on something simple and worry about things that really matter.
Dither is just hiss, and hiss is your friend.
Whereas quantisation distortion is one of the most unnatural, irritating, insidious, digital-in-a-bad-way forms of audio nastiness you can inflict on your music.
Don’t do it it.
When in doubt, dither.
A few people have questioned me about this advice, but I stand by it – check out the update at the end of this post for more discussion.
Image from Wikipedia
No related posts.