Dither is one of the least understood topics when it comes to mixing and mastering, but it can actually come in handy. We’ve compiled a one-stop shop for understanding when, why and how to use dither.
By Mike Levine
Dither is one of the least understood words in the music-production lexicon, but it’s an essential ingredient when working with digital audio, and one that you really should understand if you’re involved in mixing or mastering.
So, what is dither? It’s a form of low-level noise that is intentionally added to a digital audio file as it’s rendered to a lower bit depth. The concept of dithering might seem counterintuitive, but it’s an effective process. Dither noise actually masks what’s called “quantization distortion,” which causes noise and artifacts in digital audio. Before we go into the specifics, it’s useful to quickly review some basic concepts about digital audio.
Sample Rate and Bit Depth
When audio is digitized by an analog to digital converter (ADC)—say, when you record into your interface—and it gets sent to your DAW, the analog audio signal from your microphone or DI is transformed into a series of ones and zeros that digitally represent your audio. To accomplish that, the ADC captures snapshots (aka “samples”) of your audio at a specified rate and size, which are referred to as the sampling rate and the bit depth.
The sampling rate governs how many samples your converter captures in a second to digitize the audio, and is critical in terms of the frequency response. The general rule, based on the Nyquist Theorem, is that you must have a sampling rate that’s greater than or equal to twice the highest frequency of the audio. Otherwise, distortion called “aliasing” can mar the sound.
To capture frequencies up to 20kHz, which is the high range of human hearing, the lowest sampling rate commonly used is 44.1kHz (the rate in the CD spec), but these days, rates of 48kHz or higher are also common. The higher the sampling rate, the greater the precision for reproducing the frequency range.
The bit depth (also referred to as “word length”) refers to the size of each sample, in terms of the number of zeros and ones it can contain. It quantifies how precisely the amplitude, and therefore the dynamic range—which in this case can be described as the difference between loudest sounds that can be captured and the noise floor—will be represented when digitized.
The higher the bit-depth, the more precise the description of the amplitude will be. When you compare 16 and 24-bit audio, the difference in resolution is pretty striking. Sixteen-bit audio can represent up to 65,536 discrete levels of amplitude, whereas 24-bit has the capability for 16,777,216 levels. In terms of dynamic range, each bit is equivalent to 6dB. In 16-bit, that gives you a 96dB (16x6). In 24-bit it’s 144dB (24x6).
When you convert a file from a 24 to 16-bit resolution, your digital audio software removes (“truncates”) the last 8 bits and has to round out the data to fit the smaller number of digits available. This process is known as “quantization.” It’s a different process than when you quantize notes in MIDI, but it is similar in concept. In a MIDI sequence, when you take a sixteenth note phrase and quantize it to eighth notes, your DAW has to reassign any sixteenth notes that are not already falling on an eighth note subdivision to the nearest one. Thus, the musical resolution is reduced.
When your DAW or audio editor quantizes the bits when converting from 24-bit audio to 16-bit audio (or from 32-bit fixed point to 24-bit, or any other such reduction), there are significantly fewer steps available to map the amplitude levels to. As a result, the rounding can cause something called quantization error (aka “quantization distortion” or “truncation distortion”), which manifests itself as low-level noise or distortion.
Your music usually masks it, but with 16-bit audio, you can sometimes hear quantization error on quiet passages, or when the music is fading in or out. The bottom line is that it’s a form of distortion and you don’t want it to be present in your music. Some engineers say that even if you don’t actually hear quantization distortion, it makes the music sound harsher overall.
Dither to the Rescue
The solution is to add dither when you save the file to a lower sampling rate. When dither is added to audio with quantization distortion, it masks it (a process known as "decorrelation"), making it more random and therefore harder for your ears to discern. Instead of sounding harsh and grating, it turns the quantization distortion into a steady, low-level, analog-like hiss.
Another term you’ll see in software and plugins that offer dithering utilities is “noise-shaping.” That refers to what’s essentially EQ for dither, which is designed to change the frequency characteristics of dither noise so that it’s in a range that’s least audible to the human ear.
Because limiters are typically the last stage in the mastering chain before dithering, most mastering-quality limiters have built-in utilities for adding dither. The Waves Limiters all feature IDR technology, which stands for Increased Digital Resolution. It lets you add dither for a variety of bit-depth targets. You can choose from two different dither types.
Type 1 is designed to remove any quantization distortion from your audio. It’s comprised of dither noise that’s modified by noise-shaping. Waves recommends it for 16-bit and 20-bit processing. By adding the Type 1 dither, the 16-bit master that you’re bouncing down to, will sound like it’s actually 19-bit. In other words, you’re gaining three bits of perceived resolution thanks to the IDR dither.
Type 2 adds less noise to your signal but doesn’t completely eliminate the quantization distortion the way Type 1 does. So, the choice between Type 1 and Type 2 comes down to whether you want to completely eliminate distortion or get rid of most of it while adding less noise.
The Shape of Noise
IDR also lets you choose between three different noise-shaping options for modifying either type of dither: Moderate, Normal and Ultra. Waves recommends Normal for most situations and Ultra only for the final stage of mastering a file. You can audition the different dither and noise-shaping types by listening to their effect during a quiet section of your file, such as a fade-out or a stop where the reverb tail is ringing out alone.
If it’s not practical or you don’t have time to audition the different options, Waves offers the following suggestions: Use Type 1 with Normal noise-shaping for CD mastering. Use Type 2 with Ultra to add the least amount of noise to files 16-bit and greater.
Type 2 also supports “auto-blacking” which won’t add dither to any silent sections of the audio. If your audio does have such a section, choose Type 2. To get the maximum resolution, choose Type 1 with Ultra.
Whether to Dither?
When to add dither is a subject of some confusion and even controversy. The simplest way to look at it is that you should always dither when going down in bit-depth. So, if you’re going from 24-bit to 16-bit, you should dither. If you’re going from 32-bit fixed point (not floating point) to 24- or 16-bit, you should dither.
However, if you’re bouncing your mix to a data-compression codec like MP3 or AAC, dithering is not necessary. They’re both formats that introduce artifacts into the signal that dither won’t remedy.
The best thing you can do when encoding lossy codecs like MP3 and AAC is to use the highest bit-rate you can and still achieve the file size that’s required for your streaming platform. Don’t confuse bit-rate with bit-depth. Bit-rate measures transmission speed for streaming. You’ll most often see it expressed in kbps (kilobits-per-second) or mbps (megabits-per-second). The higher the bit-rate, the better the quality, and the larger the file size.
Hopefully, this article has helped demystify dither for you. If it still seems a little intimidating, here are some simple guidelines to follow for when and how to dither:
Want more on mastering and limiters? Gets 6 tips for limiting during mastering here.
Want to get more tips straight to your inbox? Subscribe to our newsletter here.