To dither or not to dither that is the question…
The real question is how far down the rabbit hole do you want to go? I’m sure there are plenty of incredibly talented engineers that produce exceptional sounding music that doesn’t know the first thing about dither. Dither isn’t going to make or break you and even the most technical engineers will tell you that when working in 24-bit fixed point, the majority of people aren’t going to hear a difference. But the difference is there and in our world, we are trying to accumulate small wins that will increase the quality of our work. Individually these changes don’t make a huge difference but as they accumulate with other small wins it can start to make a dramatic difference and ultimately becomes what sets a good engineer from a great one.
Unfortunately, there is a lot of misinformation out there on dither. It’s much more difficult than it should be to figure out what the hell dither is, if you should use it, and why. I like articles that get straight to the point and can sum up the topic in a practical way, quickly, easily, and without getting too complicated. This is how I prefer to look at the technical details of audio.
The truth is there is nothing I’d rather talk about less than the technicalities of digital audio, file formats, truncating bits, zeros, ones and the rest of that nonsense. Unfortunately, as an audio engineer and music producer we have a responsibility to learn new technology if there is a chance it can improve the quality of our productions.
What is dither?
Dither is a specific type of low-level noise that is added when converting the bit depth of an audio file in order to reduce quantization distortion.
Why do you use dither?
Dither is required when reducing the number of bits in an audio file to help mask any quantization errors. Dither works by randomizing the quantization errors; the added noise has the effect of spreading the errors across the audio spectrum which in turn makes them less noticeable.
If you’re converting a file from 24-bit to 16-bit you need to get rid of or truncate the 8 extra bits of information. The truncation of bits causes quantization distortion which can cause the audio to sound brittle, and gritty, as well as shrink the stereo image. A music file that was correctly dithered compared to one that wasn’t is night and day, as the correctly dithered version will always sound better, even to untrained ears.
To hear what truncation distortion sounds like you can watch Ian Shepards video “Dither or distort? Listen and decide for yourself”
Should you be using dither?
A lot of people say only dither once, dither is noise and we shouldn’t be constantly adding noise to our audio right? Well, accomplished mastering engineer Ian Shepard of http://productionadvice.co.uk and The Mastering Show Podcast says “After all, it’s just noise – and very quiet noise, at that.” Quiet noise never hurt anybody but hearing the effects of quantization errors can be quite jarring and painful. In order to minimize quantization errors that are introduced when converting fixed bit depth audio files, you should use dither every time you process a file in your DAW (more on this later).
All modern DAWs process and calculate in 32-bit floating regardless of the source bit depth. This means that bit depth conversions are happening every time you process the audio in any way like freezing tracks, bouncing in place and consolidating regions. To my knowledge, most if not all DAWs are not adding dither automatically, which means that bits are being truncated every time these processes are happening. You can prevent any conversion or need to dither entirely by always working in 32-bit floating. I would even recommend sending a 32-bit floating wav file to your mastering engineer and leave the dithering to them. You won’t need to dither until you finalize your master into a fixed point bit depth.
Even though when working at 24 bits the audible artifacts of quantization errors are negligible, by working in 32 bit floating we can eliminate the need to add dither during mixing and leave the one-time decision to be done in mastering.