Dithering is one of the most misunderstood aspects of digital audio. While it may not be as popular of a tool as your trusty EQ or compressor, dithering plug-ins are essential for producing professional-sounding tracks—especially if you plan on streaming. In this blog, you'll learn about how dither works and how to use it while mastering.
What Is Dither, Anyway?
Here's the short and sweet answer: dithering is the process of adding noise to a signal in order to reduce quantization distortion when reducing the bit depth of a file. Dithering should only be applied during the mastering stage when bouncing from 24 or 32-bit down to 16-bits. This helps ensure that every detail of your mix remains intact while streaming.
Pretty simple, right? Then why is dither such a misunderstood topic? Well, the details are a bit more complicated.
Understanding Digital Audio
In order to understand dither, you'll need a working knowledge of digital audio. Let's start with the basics.
When creating a digital audio file, you're capturing a digital representation of an analog signal. Kind of like a snapshot of sound. Much like a video with thousands of stills played one after another, digital audio is simply a series of audio snapshots played in succession, creating the illusion of sound moving through time.
In order to accurately recreate sound waves, we need to capture detailed snapshots that quantify the frequency response and amplitude of the sound. These snapshots are called samples, and they're the foundation of digital audio.
The sample rate of digital recording dictates how many samples or audio snapshots the recording captures per second—typically 44,100 or 48,000, depending on the sample rate you use.
Bit depth controls how accurately each snapshot or sample captures the details of the original analog signal. Each additional bit increases the resolution that your interface has available to measure the level of the incoming signal. Recordings with larger bit depths are capable of greater dynamic range, while recordings with low bit depths have a limited dynamic range.
Each sample or snapshot is assigned an amplitude, like a dot on a graph. However, when trying to quantify such an infinitely variable number using a finite set of values, things don't always line up.
When the amplitude of an analog signal is slightly above or below the nearest digital value, the amplitude will be rounded up or down to compensate. This causes what are known as quantization errors, which can create unwanted distortion.
It's kind of like a "connect the dots" puzzle. With only a few dots, the lines are hard and jagged, making the drawing look like a clunky version of the original. But with many dots, you can capture more of the subtle nuances of the original.
Working at a low bit depth is a kind of like carving a sculpture with a saw when you need a scalpel. It doesn't allow you to capture enough detail to accurately recreate the original source.
Why Is Dither Important?
Digital audio has come a long way since its inception. Today, most systems are capable of recording digital audio with a 32-bit depth, rendering any quantization distortion practically inaudible.
However, even if you record at the highest possible bit rate, you still run the risk of creating quantization errors when bouncing your track down to a 16 or 24-bit file. At 16-bits, quantization distortion becomes more audible—especially during quiet passages or fade outs.
When the number of bits is reduced, so is the available number of discrete amplitude values, resulting in more quantization errors. Thankfully, there's a way to prevent these errors. That's where dither comes in.
Back in the early days of digital audio, some observant engineers discovered that mechanical aircraft computers performed more accurately in flight than on the ground. It turns out that the vibration from the plane's engines actually helped keep the moving parts inside of the computer from sticking, resulting in greater accuracy.
Dither works in much the same way. It adds a layer of randomized noise to the track to better assess the amplitude of each sample, preserving the original details of the recording.
The key here is that the noise needs to be completely unrelated or "de-correlated" from the original signal. When applied correctly, the system is able to round amplitude values more accurately, which helps preserve the signal and removes any added distortion.
How Does Dither Work?
Now that you understand when and why to use dither, let's talk about how it actually works.
When reducing the bit depth of an audio recording, all of the information from the original 24 or 32-bit file file needs to fit into a 16-bit file. In order to do this, a lot of the information is lost.
In the most basic systems, any amplitude values that cannot be replicated are simply cut off, or truncated, causing unwanted distortion. More sophisticated systems will round amplitude values up or down, but they're not 100% accurate. Rounding errors also result in unwanted distortion.
Say there's an amplitude value that needs to be rounded, but it's directly in the middle of the closest available values. There's a 50/50 chance that the computer will make the right choice. Instead of simply guessing, like a human would, it always makes the same decision—either up, or down.
Dither plug-ins add noise to the signal, which essentially randomizes your computer's decision to round up or down. Dither plug-ins prevent truncating and rounding errors and minimize unwanted distortion when lowering the bit-depth of a track.
When To Use Dither
I remember when I first started mixing, I used to put a dither plug-in on the mix bus because I could swear that I heard it making a difference. Spoiler alert—it did not.
Dither only needs to be applied when reducing the bit depth of a file. Typically, this occurs during the mastering stage, when bouncing files from a 24-bit or 32-bit session down to 16-bits (or lower, if that's your thing, but most streaming services require at least 16-bits). This is the only time you should apply dither.
Any time you change the sample rate or bit depth of a recording, you run the risk of introducing distortion, which is why it's recommended that you only do so when absolutely necessary. Generally speaking, you'll only need to apply dither when bouncing a file outside of your DAW, which is why most DAWs offer a few different built-in options (more on that later).
However, it's important to note, if you're bouncing a mix to send in for mastering, you should send a full 32-bit file. The mastering engineer will apply dither upon bouncing the final mastered track. Sending your mastering engineer a 16-bit file limits their abilities when trying to fine-tune the dynamics, and can cause unwanted distortion when working with 32-bit plug-ins.
Dither plug-ins exist for those who like to bounce and store files within a session, and offer additional flexibility over built-in dither options. If you are using a dither plug-in, it should always be the very last plug-in in your signal chain, which is why many limiters feature built-in dithering capabilities.
When Not to Use Dithering
Remember, dither technically adds noise to your signal, so you should avoid using it unless absolutely necessary. You only need to use dither when bouncing a file at a lower bit-depth than it was recorded. You should always apply dither when bouncing a 16-bit file from a 24 or 32-bit mix.
Technically, you do not need to apply dither when bouncing a 24-bit file from a 32-bit float session, as it has the same bit depth. However, you should apply dither when bouncing a 24-bit file from a 32-bit fixed point mix. Having said that, you probably won't run across this situation very frequently, as most streaming services and CD manufacturers require 16-bit files.
Generally speaking, you should avoid applying dither more than once. However, if you're working with a 32-bit fixed point session and need to bounce down to 24-bits so you can use some 24-bit recordings in your session, you should apply dither. Then, when bouncing the final 16-bit file, you'll apply another round of dither.
Different Types of Dither
In 1997, a set of audio engineers from Dolby Labs, Weiss Engineering, Millennia Media and Z-Systems set out to create "the most sonically transparent dithering algorithm possible." The following year, they unveiled a set of commercial dithering and noise shaping algorithms known as Psychoacoustically Optimized Wordlength Reduction or POW-r.
Today, all of the leading DAWs (including Logic, Pro Tools and Ableton) feature built-in dithering processors based on the original POW-r algorithm. In fact, most DAWs offer multiple dither options based on the POW-r algorithm. While the differences are generally subtle, each algorithm is optimized for different types of music.
But before we get into the different types of dithering algorithms, we need to talk about Noise Shaping. Noise shaping is essentially the process of applying EQ to dither noise to make it less audible. At 24-bits, the noise created by dither is generally inaudible, but noise shaping can make a noticeable difference when bouncing files down to 16-bits (or 8-bits or those rare Chiptune sets).
Next, let's take a quick look at what makes each of the POW-r algorithms unique.
Type 1: Type 1 POW-r dithering is typically used for loud mixes with low dynamic range, like a highly-compressed rock or pop song. Type 1 dither does not use noise shaping, meaning there is a flat frequency response with no EQ curve, making it ideal for bouncing high-res 24-bit files.
Type 2: Type 2 POW-r dithering is typically used for speech. It uses a simple noise shaping curve with a dip around 2 kHz and a boost above 14 kHz. This helps the vocals remain audible while still preventing quantization distortion.
Type 3: Type 3 POW-r dithering is typically used for highly dynamic recordings, such as orchestral music or film scores. It also uses noise shaping, but features a much more aggressive EQ curve to further eliminate quantization distortion.
Now that you're a certified dithering expert, you can use these techniques during your next session to make sure your mix is free of errors and unwanted distortion.