Can You Hear the Effects of Dithering?

Certain topics are pretty much guaranteed to light up internet forums, like Mac vs. PC, Les Paul vs. Strat, and dither vs. well, different kinds of dither, and whether dither is even necessary — or some scam perpetrated on us by audio PhDs who don’t live in the real world. And like so many esoteric audio concepts, the answer is: it depends.
You can go deep down into technical rabbit holes with dither, so we’ll do more of an overview. Most importantly, we’ll describe how to evaluate dither objectively and decide what sounds best with your music.

What Is Dither, Anyway?
Today’s hard disk recorders most likely have an audio engine with at least 32-bit, floating-point resolution. But currently, although there are some high-resolution audio formats intended for consumers, your mix usually ends up streaming online in MP3 or AAC format, or in what’s still the most common physical delivery medium — a 16-bit/44.1kHz CD.
Dither — the noise added to an audio signal prior to quantization or word-length reduction to minimize the resulting distortion and noise
Before dithering, an audio file’s “extra” bits (for CDs it was anything beyond 16-bits) were simply truncated (discarded) when creating the CD master. This meant that, for example, decay tails below the 16-bit resolution limit just stopped abruptly. Maybe you’ve heard a buzzing sound at the very end of a fade-out or reverb tail; that’s the sound of quantization noise, which occurs because the least significant bits can’t reproduce smooth audio variations. Dithering is a process that adds a controlled type of noise to the audio signal, which can reduce (or even eliminate) quantization noise at the lowest levels (fig. 1).
Figure 1: Both waveforms were set to a peak of -80dB, exported, then imported and amplified back up so the peaks hit 0dB. The upper waveform has no dithering and sounds spiky and distorted. The lower waveform has dithering applied. It looks and sounds smoother.
It may seem odd that adding noise can improve the sound, but psycho-acoustics is on our side. Any noise added by the dithering process has a constant level and frequency content, so our ears have an easy time picking out the content (signal) from the noise.
This article’s focus is on understanding how dither affects your audio. Note that we’ll reference audio examples when appropriate, but individual files are included for loading into your DAW to make for easy comparisons.
Experiment #1: Can You Really Hear Low-level Audio?
This is an experiment to do in your DAW, so download the audio WAV files by clicking here.

Load the stereo 24/44.1 Reference File.wav into your DAW, and set the volume for a comfortable listening level.
Normalize the file to -60dB without changing the listening level.

Unless your idea of a comfortable listening level is to blow your ears off, playing the reference file back at -60dB will be inaudible or at least very faint. Bear in mind that this is the kind of level (and lower) where dithering does its thing.
As a result, dithering has the most impact with music that has wide dynamics, doesn’t necessarily hit 0 very often (if at all), and is at a fairly loud playback level. For example, you won’t hear the impact of dithering on pop music that’s always kissing 0dB, but you may hear it on the reverb tails of acoustic symphonic recordings playing back from 16-bit CDs.

Experiment #2: Quantization Noise with 16-bit Audio
Let’s hear what quantization noise sounds like.

Load the file -80 no dither 16-bit normalized.wav. This file is the result of normalizing the 24-bit reference file down to -80dB, exporting it as a 16-bit file with no dithering, and then normalizing it to full volume to hear the actual quantization noise.
Play back the file. It should sound like your worst audio nightmare — that’s the quantization noise.

-80 no dither 16-bit normalized.wav

https://www.sweetwater.com/insync/media/2018/12/80-no-dither-16-bit-normalized.wav

Load the file -80 no dither 24-bit normalized.wav.This file was also normalized to -80dB, exported with no dithering, and normalized to full volume.
Play back the file. It should sound very much like the reference file, because its lower 8 bits weren’t truncated, so there’s no audible quantization noise.

-80 no dither 24-bit normalized.wav
https://www.sweetwater.com/insync/media/2018/12/80-no-dither-24-bit-normalized.wav
This experiment also brings up an interesting side note about 24-bit/96kHz “high-resolution” recordings: Maybe it’s not the high sample rate that causes people to think this format sounds better, but the 24-bit resolution.
Experiment #3: What Dithering Sounds Like
Although we’ll be applying dithering to extremely low-level signals, we’ll normalize them to 0dB in order to hear the dithering in a really obvious way. (But remember how in Experiment #1, -60dB is a really low level, and these examples use audio that’s even 20dB quieter — so, of course, the noise will be prominent.) The following files are all 16-bit.
-80 Triangular.wav
https://www.sweetwater.com/insync/media/2018/12/80-Triangular.wav
-80 Triangular Studio One.wav
https://www.sweetwater.com/insync/media/2018/12/80-Triangular-Studio-One.wav
-80 Pow-r 1.wav
https://www.sweetwater.com/insync/media/2018/12/80-Pow-r-1.wav
-80 Pow-r 2.wav
https://www.sweetwater.com/insync/media/2018/12/80-Pow-r-2.wav
-80 Pow-r 3.wav
https://www.sweetwater.com/insync/media/2018/12/80-Pow-r-3.wav
-80 Waves Type 2 Ultra
https://www.sweetwater.com/insync/media/2018/12/80-Waves-Type-2-Ultra.wav

Load these files (that you downloaded) into your DAW. The Pow-r dither algorithms are used in several DAWs; the Waves algorithm is included in the L3 Multimaximizer and other mastering-oriented plug-ins.

-80 Triangular.wav
-80 Triangular Studio One.wav
-80 Pow-r 1.wav
-80 Pow-r 2.wav
-80 Pow-r 3.wav
-80 Waves Type 2 Ultra

If available, enable Exclusive Solo (called X-OR Solo in Pro Tools) in your DAW so it’s easy to switch among the various files for comparison.

Triangular demonstrates a generic, wideband noise signal. It helps get rid of the horrible quantization noise by replacing it with smooth noise. Granted the noise sounds loud, but again, this is happening at a very low level.
Triangular Studio One also features a wideband noise signal, but the algorithm Studio One uses makes it less annoying than the more generic version.
Pow-r 1 is designed for low dynamic range content.
Pow-r 2 shifts the noise into a higher part of the audio spectrum.
Pow-r 3 is considered the most transparent of the three options. The three Pow-r options all use noise shaping that shifts the noise to less audible parts of the audio spectrum.
Waves Type 2 Ultra from Waves is a good example of a “designer” dithering algorithm that’s remarkably transparent.

Note that not only DAWs, but plug-ins (like the Waves L series mentioned previously) and other mastering-oriented processors offer dithering options (fig. 2).
Figure 2: IK Multimedia’s T-RackS One processor offers four types of dither. The dithering options become available if a file is being converted to a lower-bit resolution (e.g., 16 bits).
Which Flavor of Dither Should You Use?
In the examples above, I’m pretty sure you’ll agree the hiss is an improvement over the quantization noise. Although people may not hear a noticeable difference at such low levels with most music, our ears are pretty remarkable transducers, and anything we can do to make audio sound better is worth it.
Different manufacturers use different noise-shaping algorithms; the differences are usually subtle. Sometimes you’ll have a choice of dithering and noise-shaping algorithms so you can choose the combination that works best for specific types of program material. Not all these algorithms are created equal, nor do they sound equal. However, I recommend not obsessing over this — I think your time is better spent obsessing over chord progressions and lyrics.
To determine how dithering sounds with your music:

Normalize a copy of your final, 24-bit (or higher) mix down to a low level, like -70dB to -80dB.
Export it to multiple 16-bit files, using various dithering options.
Load the dithered, 16-bit test files into your DAW.
Normalize them up to maximum so you can hear the results of the dithering.
As you master your mix, apply the one you like best.

Cautions!
Because there are conditions under which dithering makes an improvement, you might as well add it as a matter of course. However, there are a few cautions.

Apply dithering only when converting a high-bit-resolution source format to one with lower resolution, and apply it only once. Typically, this is from your high-resolution master or mix to a 16-bit, mixed-for-CD format. In particular, applying noise-shaped dither more than once is asking for trouble.
If you give a mastering or duplication facility two dithered 16-bit files to be crossfaded, note that crossfading the dithered sections could lead to artifacts; it’s better to crossfade the two files prior to dithering, and then dither the combination.
Check any programs you use to see if dithering is enabled by default, or enabled and saved as a preference. In general, you want to leave dithering off, and enable it only when needed.
Check whether two plug-ins are dithering at the same time. You want dithering to be the signal chain’s last processor, and preferably, be post-master fader. Don’t double up on dithering.
Dithering a master file is appropriate for converting something like a 24-bit file to MP3, but it’s important to use basic dithering options like triangular or rectangular. Noise-shaped dithering can interfere with the data compression process.

And there you have it — what you need to know about dithering. If your music includes wide, natural dynamics, proper dithering can indeed give a sweeter, smoother sound free of digital quantization distortion when you downsize to 16 bits.
The post Can You Hear the Effects of Dithering? appeared first on inSync.