OK...this is a topic we started in the middle of another thread elsewhere here on RO, and I think it deserves a thread of its own. I was going to post it under the pluginsforum, but felt that it would get more attention here.
In another thread, member angrynote (member #5795)wrote:
----
"I have just started exporting in 32 bit float file with whichever sample rate I'm using and then exporting into Sound Forge's dither plugin for maximum fidelity when going down to 16bit. Much prefer it to the UV22 plugin because it maintains the headroom in the mixes whereas the Apogee dither in Cubase kinda kills the peaks and squashes them."
----
I responded:
"I have always used the UV22 dithering scheme without giving it another thought because I have always heard it was much superior to everything else. I also own Sound Forge (and Wavelab), but have always gone back to Cubase for the dithering from 32-bit back down to 16-bit. I guess now I should probably try it both ways and compare the difference... "
angrynote came back with:
----
"My opinions are only that in situations where I am trying to maintain the highest level without digital distortion or other unpleasant noises - I find that the 32bit export is wonderful because it allows the track to breathe as it does in the Cubase environment. There are a few issues however - if there are too many loud peaks in any file exceeding sometimes 5db then you will suffer some sort of artifacts in your mix.
Try a loud track that is clearly peaking above zero but sounding fine to your ears - mix it in 16bit with the UV22 and you'll hear that although the track sounds fine - it can't breathe at the top. Sounds strangled - but again that's just me -
The Soundforge 5 dither features high pass triangular and noise shaping options that when converted down to 16bit - not only take seconds but sound wonderful and it still breathes.
This all came from my anti-mastering plugins within software - mastering engineers are trained professionals and we cannot expect to maximise our track without problems. So I try to record everything HOT - almost as if I'm compressing it for CD release - and there I have the high levels. "
----
Now since these couple of posts, I have been reading in my Waves manual about their dithering process (that is used with the L1 plugin among other places). They call theirs IDR (Increased Digital Resolution)developed with the late Michael Gerzon. Waves of course touts their dithering scheme as being the best there is, and they give all sorts of technical reasons why.
Before I create a whole book here, I would like to open this topic up for discussion. I am pretty ignorant here, as I mentioned above, I have always just used the UV22 scheme that now ships with Cubase (and will be in the new Wavelab 4.0 release, btw), and never really thought about it.
As always, our ears should be the deciding factor, but I am anxious to hear other comments.
Comments
A good way to think of dither is that it is a method of randomiz
A good way to think of dither is that it is a method of randomizing the inevitable distortion from wordlength reduction into something that sounds like noise. It is NOT just a masking signal as way too many software manuals, magazine articles and books ignorantly suggest. I've even run into software engineers who were taught this masking noise nonsense in college.
Dither becomes a tricky subject because you often CAN'T just use your ears! This is because most DAT machines and DAWs never actually reproduce a 16 bit signal until you play your final 16 bit tape or file back. When they are in "input," you are hearing a 24 bit signal reproduced by an 18, 20 or 24 bit D to A converter. Hence the only way you can compare dither in this situation is by playing back two versions with the different dithers. This is why many people (including the occasional magazine writer) claim they don't like using any dither. A dithered signal SHOULD sound worse until you go back and play it. Then it will usually sound warmer and larger than not dithering.
As a history note, in the early days dither was never an option. The high-end gear used it and the cheap stuff didn't. It really shouldn't ever be a user selectable option. It's like having optional bias on an analog recorder that defaults to "off." The only reason we are even talking about dither is because it got tacked on to a number of applications after people complained instead of having been designed into DAW systems from the start. The appropriate user interface would be an output word-length selector perhaps with a dither flavor adjustment on really "tweak" gear. Equipment or software that does not AUTOMATICAlLY dither all wordlength reductions and not dither cases where there is no reduction is just plain poorly designed.