Lengthening words

Author's writing guides advise caution when it comes to the number of words in a sentence. Fewer are preferred. Likewise, dollar words should be avoided when pennies will do. The same isn't true for digital audio, where longer is better. We prefer 32 bits to 16—more information rather than less. Twenty years ago when Bob Stadtherr, Arnie Nudell, and I designed the Digital Lens there wasn't media available with more than 16-bits. Despite the lack of media, a growing number of DACs began accepting longer word lengths in anticipation of higher resolution media: 18, 20, 22, and a few at 24. We provided buttons on the remote to extend 16-bit words to as long as 20. How did we do that without access to the original higher resolution media? We added noise. Or, in digital parlance, dither. When you add dither to extend word length there is no new information, yet it can sound better. How was that possible? In those early days, the differences in sound were attributed to the use of different digital filter algorithms in the DAC. 16-bit words had one set of filter parameters, 18-bit another, and so on. There was no added musical information but from the user's standpoint, it didn't matter. There's more to the story than just accessing different algorithms. Dither, when used in the A/D process, or when encoding and decoding, can actually add detail—filling in the spaces between digital steps—and it can do that without knowing what was in the original. Here's an example: The first image on the left is a very low resolution, low bit depth image. Note the blocky colors. The image to its right is the identical image but with dither added. Note how much smoother the image is. This is because dither is filling in the gaps. These same techniques applied to audio give a reasonable approximation of more detail. Is dither a panacea to high-resolution audio on a bandwidth budget? No, but it can certainly smooth rough edges. Sometimes, noise isn't such a bad thing.