 |
 |
 |
 |
 |
 |
|
|
|
December, 2006
 
MySpace and Hi-res, Nuance and Dither
An Oxymoron? by Jonathan Wyner "The consumer is listening to hyped radio broadcast and singles on Ipods on earbuds which is like, the end of the world for fidelity and why the heck should I care about 96k/24bit 'cause it still is gonna end up as a 96kbps MP3 on myspace!" This is the argument I hear all the time from engineers and consumers. Two months ago I wrote a piece about recording and mixing using high resolution systems and how it's so easy to do it and there's no excuse to not do it, but looking back at the first line above it might still seem like it's hard to justify. Perhaps we can take a lesson from understanding dither to illustrate why we should care about the resolution of the master in the face of low-res delivery. So what is dither anyway? Roughly speaking, it is noise added to the signal you are creating that is equal to 1/2 of the dynamic range of the least significant bit of the audio. Each bit in a PCM digital audio system describes 6db, and the theoretical dynamic range of a 16-bit CD is 96db and the level of noise/dither added would be about 3db, but why add noise? There are two reasons/benefits. First of all, it enhances the performance of the D to A converter in the following manner. When a 16-bit converter reproduces signals that are near silence, given that it has a finite ability to reproduce audio (6db per bit), at some point it will transition from on to off and back again. In other words, when the signal falls below the level that can be reproduced by the converter, it shuts off. That transition from quiet to off is abrupt. Digital audio is an either/or proposition to some degree, a zero or a one. If the number is too small, it is rounded down to zero. That transition from one to zero and back again can sound bad, jagged, edgy and distorted. That's called quantization noise. Distortion. If you play a 24-bit file back on a 16-bit converter without dither, the name for this toggling on and off is actually called truncation distortion. However, adding dither prevents the D to A converter from ever turning completely off, so while you have added noise to the signal, it is generally agreed that the noise is way better than the sound of the distortion. If you have never heard the sound of quantization distortion, you can by running a signal through a program such as Bit-Crusher at a resolution of 1 bit and listen. (Turn your monitors DOWN before you do; it sounds nasty!) What you'll hear is the sound of what happens to the quiet part of your program when you truncate. The sound might work on a drum loop in a techno tune, but it's generally not the sound you are looking for in your recording! The second benefit of dither is that you can actually hear program BELOW the noise floor when you properly add dither. Obviously, a 24-bit file contains a lot (8 bits worth) of quiet signal that a 16-bit system cannot capture. It's just too quiet. However, if you add noise to the 24-bit file, the noise gets mixed with the quietest information and when that happens, SOME of the quietest information is now raised so it is within the 96db of dynamic range and can be captured by 16 bits of conversion! It's so simple and yet it's like magic. Dither only gets you this benefit when it is applied to a file before reducing the bit depth—in other words, going from 24 bits to 16 bits. If you apply noise to a file that's already at 16 bits there is no additional low level information to recover. OK, so, if dither can help you effectively get more than 96db of the dynamic range of a good recording into a 16-bit signal, do MP3's work the same way? The answer is yes. Of course, anytime one goes from a higher resolution signal to a lower resolution signal some information has to get thrown out, but what I want to point out is, having the extra information available before reducing a file to a lower-res target will yield a better result. There is a law of diminished returns and you could argue that 384KHz doesn't yield enormous benefit over 96kHz when your target is 96kbps MP3, and the trade-off of taxing the dsp's and storage in a system with the extra speed and file size generated might not be justified. But I think you can see that to an extent, some of that hit IS justified, no matter how low-res your end use is … that is assuming we're dealing with a music application. Telephony is another story all together. Jonathan Wyner has mastered more than 4000 CDs during the last 21 years, spanning every musical idiom (and some nonmusical idioms as well!). He is an adjunct professor at Berklee College of Music in Boston, Mass. Credits range from the extremely well known (James Taylor, David Bowie, Aerosmith, Kiri Te Kanawa) to the more idiosyncratic and independent artists/labels. Jonathan prides himself on the fact that he is among the elite group of mastering engineers to have run a marathon in under three hours repeatedly. Visit www.m-works.com. |