Dithering at the mastering stage?

[Rad]

I have a question about the possibly (in)appropriate use of dither in the mastering stage.

Many effects and mastering limiters, such as the Waves L-series, have several dithering options - Type1, Type2, Normal, or Off. While it is known that anytime when there is bit reduction, say, from 24 to 16 bits, dither should absolutely and always be used, it is much less clear to me what is the appropriate setting if I just want to maximize a 16-bit file without any further bit reduction. Should the dither option be left on or off?

From the literature, it would appear that all plugins that process the signal internally at more than 24 bits, need some kind of dithering to get back to 24 bits, so here's one dithering stage. Then while converting this file down to 16 bits, there's one more dithering stage. Thus, I'm a bit unclear about what exactly switching dither on or off does, because I can't seem to be able to figure out which of these two dithering stages they are talking about.

[Tarktones]

It's hard to say. I'd check the manual on the specific plugin and if you can't find a good answer, e-mail the OEM. I have a feeling it changes from plugin to plugin.

[DAS]

Most manufacturers are still pretty vague about where and how much dither they are applying in their plug-ins. Not being them I can't be sure of all the reasons why, but one complication I do know exists is the trouble between how things work in a fixed point versus a floating point system. Changing those dither selections on a plug running on a fixed point processor may be substantially different from what happens on a plug in a floating point processor.

The way the Waves thing was always explained to me is that the dither selection should be made primarily based on whether the file in question would undergo subsequent processing. You can find more detail about that in the manual. Of course this doesn't exactly answer your question, but it is an answer you can work with.

[dpd]

anytime the bit depth is lowered, dither should be applied - internal to a plugin when sending the final processed output down from its internal resolution to its output bit depth (e.g. 24 or 16) and whenever you bounce a recording to its final bit depth.

[Tarktones]

I think the question is wether or not the dither on a number of plugins is specific to internal processing or if it's intended as the last step in the chain.

i.e. if I don't have Ozone 3 as the last thing in the chain, should I dither to 24 bit since it processes at 64 bit and leave the 16 bit dithering to whatever the last plugin is, or should I leave the dither off completely since I'd assume the plugin would automatically dither to the audio file's base resolution?

[Audioholic]

If a plug in process at a certain bit rate, say it advertises 48 bit operation, it doesn't not make your work length now 48 bits. If you used the plug in last, and bounce down, your DAW would still output whatever length you recorded at, 16, 24. IF from 24 to 16, you would want to dither probally as your last plug in the chain, ie, L1 with dither option on the master.

the types on the L series are a little bit confusing, I think it adjusts the sharpness or something of the dithering, and is says something like its possible if further editing is gonig to be done, you will hear a pop if using type 1 dithering or so forth, who knows.

[Rad]

Right, it is confusing. What I meant was more like what happens when we start with a 16-bit file and process it with the plugin without further bit reduction - i.e., start with 16-bit and end up with 16 bits. So I guess the real question is whether by turning the plugin's dither to OFF I am switching off the internal dithering that the plugin does from 64 to 16, (which would be a mistake to do), or switching off an extra dithering stage that the plugin offers afterwards specifically to help people with bit reduction from hi-bit formats (say 24 bits) to 16.

If it's the former case, then switching off the plugin's internal dither is a capital mistake and will result in distortion. But if it's the latter, since the file is 16-bit already, leaving final stage dither off is the correct thing to do, since there's no further bit reduction and all it does is essentially adding totally unnecessary noise.

As DAS has correctly pointed out, most manufacturers are quite vague about where and if they use dither. I am currently leaning towards the opinion that at least the L-series generates dither for quantizing down, that is, "final-stage dither", but that doesn't answer the question whether the plugin has its own internal dither before that (in the light of what was said above)

[Audioholic]

I think if you start with 16 bit, you won't need to dither down, IE, can turn off the plugs dither functions, and you will be fine. Turning it off won't do anything bad, meaning even if the plug is supposed to operate at a higher internal bit rate on the plug, it still shouldn't alter your word length and hence not the need to dither.

[dpd]

Various dithering algorithms have all kinds of unique noise-shaping to achieve different types of results. Wiki (http://en.wikipedia.org/wiki/Dither)reference

[DAS]

but that doesn't answer the question whether the plugin has its own internal dither before that (in the light of what was said above)

I'll bet it does...at least when running on a fixed point system. Rad, it might not be hard to test this. Run a 16 bit file through it and set the output at 16 bits. Listen to the result at different settings. Also, you could call them to see what they say. One of their guys is going to be here next week. If I have time to connect with him about it I will.

[Rad]

Thanks, DAS. I bet you probably have more important stuff to do, but as a general knowledge it would be useful to know - not all clients provide 24-bit files and it would be useful to be sure we're not degrading the quality of their masters by either adding unnecessary noise, or alternatively, by skipping dither at a stage where it should be used.

As a matter of listening, yesterday I tried to check if there are audible differences with and without dither, and I've gotta say that if there are any, they must be very subtle. My personal impression was that with Type 1 dither with low/moderate noise shaping the material sounds a tiny bit more transparent than with either ultra shaping or Type2, but both dithered versions introduced a barely noticeable smearing of the highs relative to the no dither position. (In any case, with no dither, I could not hear any audible signs of distortion.) I take this as a sign that I probably should leave dithering off.

To be honest, it did not occur to me to check if my PC is running on a floating-point processor, but I believe it does (AMD Turion TL-58). In this case, what I heard may be due to the floating-point math, so if the results are saved in a fixed-point file like .wav it is quite possible it may sound different.

[DAS]

If you're running on a native based computer system, as opposed to dedicated DSP hardware such as Pro Tools HD or any of a number of hardware recording and mixing systems, then you can be pretty sure you're in floating point land.

[dpd]

dither to 8 bits and play with the various noise shaping to easily hear the effects.