Why do I need dither?
Bit depth represents dynamic range, with each bit adding another 6dB. A 16-bit file with peaks close to full scale (0dB) will have all 16 bits active. But if you turn it down by 6dB, the loudest bit won’t be used, and you’ll be left with only 15 bits of actual information. By the time your signal drops to around 80dB below full scale, there are only a couple of bits of resolution left: 2-bit audio sounds pretty crunchy and distorted!
Adding a touch of noise before you truncate (ie, throw away the least significant bits) randomises the resulting errors, so they’re not related to the audio signal. In practical terms, this means that reverb tails disappear smoothly into the noise floor instead of turning into crunchy distortion. The noise doesn’t just act to mask the distortion: it actually prevents it from happening, and the only side effect is the slightly higher noise floor.
For perspective, both the dither noise and the distortion it fixes are likely to be inaudibly quiet with most pop music masters – dither is more critical when mastering dynamic material such as classical orchestral recordings. But creating a 16-bit file without dither is wrong, even if you can’t hear the difference!
Fancy ‘name brand’ types of dither or options such as ‘noise shaping’ don’t provide any better performance in terms of reducing distortion. Rather, they shape the noise in various ways to make the noise itself less noticeable. But as the noise is already inaudible in most cases, these are very much optional, and any type of dither will be fine.
If your DAW has an option to dither files automatically on render, you could just turn that on and not worry about it any further! However, it has been known for bugs to creep into DAW dither algorithms, and sometimes these go unnoticed until much later in the process. So, if you want to make sure that everything is working correctly, you should definitely check out our tip at the bottom of the page for a simple testing method.