Feb. 05 2020
Dither: you've seen it in your DAW's render options, maybe you've seen it in your favourite limiter too, so what is it? The short answer: dither is noise. But why would you want to add noise to your music? To understand that let's get into a bit more detail.
An audio waveform has two components; frequency, represented on the horizontal plane of a graph, and amplitude on the vertical plane. In digital audio the frequency component is captured by the sample-rate (44.1kHz, 48kHz, 96kHz, etc.), and the amplitude component by the bit-depth (16-bit, 24-bit).
The bit-depth defines the maximum dynamic range it can store, from 0dBFS at the loudest point down to about -96dBFS for 16-bit, or -144dBFS for 24-bit (Fun Fact: the max range humans can perceive is roughly 140dB, so 24-bit has us covered).
While the peaks of your music are probably close to 0dBFS, the amplitude continuously goes up and down so even if your music is loud on average, some of the digital samples will be taken at a low-amplitude portion of the waveform.
Most of us are recording and mixing at 24-bit, but CDs and most distributors for streaming platforms still require a 16-bit file, so what happens to the sample points of a 24-bit signal that are below -96dBFS when converted to 16-bit? They all get shoved into the new lowest possible point, and the details that previously existed are lost, and the sound becomes distorted.
By adding low-level random noise (dither) to the signal, we can push those points up into the desired range without forcing them all into the same position. The result is a signal that sounds much more similar to the original 24-bit signal, but with a little bit of noise in the background.
SEE FOR YOURSELF
The same process is used in digital images as well, and you've probably seen the results of bit-reduction without dithering while watching Netflix without realizing that's what was happening. Instead of amplitude, bit-depth defines the number of different colours that can be displayed. As files get compressed to smaller file sizes (i.e. for online streaming) the range of colour gets reduced, and colours that no longer have their own point get rounded to the nearest available point. Typically shades of black are the first to go, resulting in "banding" in dark scenes - this is roughly equivalent to the quiet points in our audio.
Have a look at the following images:
The second image was saved using the lowest quality JPG option in Photoshop. You can see the gradually darkening gradient of the sky in the first image is now a series of colour 'bands' because the data points between those colours have been thrown away.
The third image had a small amount of noise added to it before being saved with the same lowest quality option. The sky's gradient much more closely resembles that of the original photo, but with a slight 'grainy' texture to it - a texture that's much easier to tune out than the banding in #2.
This is exactly what happens when we dither audio before reducing the bit-depth, the added noise is far less noticeable than the distortion that would occur if we hadn't used dither.
WHEN TO USE IT
So now that we know what it's for and how it works, when should it be used? Any time you're going to a lower bit-depth. The most common scenario will be at the very end of mastering when creating the 16-bit distrbution master. Different noise-shaping options can make the noise itself less noticeable in some cases, but may allow some distortion to still occur. Generally speaking the added noise is so quiet already that noise shaping is somewhat unnecessary, so regular everyday triangular dither is a great set-and-forget option to get rid of distortion from bit-reduction.
If there's a topic you'd like to know more about, send in your article suggestions!