Mastering und "Dither" - Perlen aus der MIO-Liste

Die MIO-Liste ist eine Mailingliste wo sich die Metric Halo User austauschen, neben Bob Katz, Ed Abott und anderen bekannten Namen trifft man dort auch z.b. Brian Willoughby - diesmal zum Thema Dithering und Mastern für Itunes...

Once you've stored audio in a file (fixed point), and are dealing with a mastering workflow, there really is no concept of "over" because the format literally cannot contain a value above 0 dBFS. Proper dither can only add or subtract 1 LSB value, so while that could technically create an "over," it would only occur for the two largest sample values, one each at the positive and negative range, in which case "clipping" is not really clipping so much as removing the dither for those samples. As you mention, you'll probably never hear any "clipping" - and I would rephrase that to say that you'll never hear that the dither was not performed exactly for those rare, individual samples at the very end of the range.

By the way, the MIO red LED clip indicators fire at -0.1 dBFS in the digital domain before the D/A conversion. Since true clipping is not possible to detect after the conversion to the 24-bit fixed format, the MIO is designed to show red when samples get "near" clipping. It might have been possible to detect true clipping, but it would have required out-of-band signaling and would probably be very complicated to manage. I only mention this because some signal flows will fire the red "clip" LED even though the audio has not literally been clipped.

Of course, when processing audio in a system like MIO or Quattro, a slightly different format is usually used, such as 80-bit SHARC float or 32-bit float. In those formats you can go over, and would need to limit the signal if you do any processing other than dither.

I'm certain that following dither with a standard limiter would ruin the dithering. Dithering must always be the last stage, and limiting counts as another stage. The only limiter algorithm that would not undo the dither would be exactly equivalent to hard clipping - which is what you get without a limiter anyway (unless there is something seriously wrong with the float-to-fixed conversion). So, yes, it's always good to leave some headroom in your audio when mastering, but I say that has nothing to do with dither.


Also, there are two schools on dithering.


A) Some say that every bit reduction must be dithered - and that would necessarily include every single bus, and nearly every effect. Pro Tools had an option for this with their dithered busses. MIO has this option, but you'll have to place dither on every bus manually - I'm not sure that's technically value, though. You can't really dither effects on the MIO because you have no access to the samples before they're truncated, and dither only works before the truncation, not after.

B) Others say that you really only need dithering at the final stage, with the assumption that the bit reduction never goes below 24-bit. Folks who prefer dithering only once still agree that any part of the work flow that's reduced to 16-bit would need dithering at that stage as well as the final, but most in this camp simply avoid reducing below 24-bit until the very final stage. These days, it's not really that difficult to maintain a 24-bit signal path from start to finish.

So, you're correct that there's reason to dither to 24-bit in the MIO, but I'd say that's only some of the time. If you plan on shipping the audio that's coming out of your MIO mix without any further processing, and you're using a high-definition 24-bit format, then I'd recommend 24-bit dither in the MIO and then don't touch the data. However, if you're going to dither that to 16-bit for CD, or you're going to deliver to Apple for "Mastered for iTunes," then I'd recommend not dithering the 24-bit data as it comes out of the MIO. In that case, dither to 16-bit for your CD master, and leave it to Apple to dither the iTunes files.

Part of the deal with Mastered for iTunes is recognizing that dither is entirely unnecessary for lossy formats because the psychoacoustic model masks quantization noise in a different way than dither, and it's not good to combine both. Technically, this creates a third school

C) on dithering - which is no dithering at all for lossy delivery formats. That said, I do dither to 24-bit in the MIO when listening to lossy audio, because the lossy reconstruction process occurs at 32-bit float, all the way from iTunes through the FireWire link to the MIO, before hitting the 24-bit converters.

I hope these comments are useful for MIO users. I use my ULN-2, 2882, and LIO-8 for recording, mixing, mastering, HD Audio playback and even casual listening. I find that it's useful to pay attention to the ramifications of formats and dithering, and use lots of bit meters and frequency analysis to keep on top of the various digital and analog stages.

Brian Willoughby
Sound Consulting

Kommentar schreiben

Kommentare: 0