In a dither about dither

[C00kie]

In practice you only need to dither while converting the mastered final version to 16bits for CD distribution.

When the seperate tracks were recorded in 16bits, you still need to dither as a final process. Because all of the seperate tracks will be mixed down (and each passing a fader at -10dB or so) to 32bits fp resolution.

When going from 32bit floating point to 24bit ints, you don't need to dither. Because 32bits fp is in fact 24bit worth of data with 8 bits for the mantissa. Yes, in theory there will be truncation errors in the process. But these are at around -144dB, so not audible at all.

[cosmicdawn]

If you edit a 16 bit wave file and resave it at 16 bit you don't add dither.
If you edit a 24 bit wave file and resave it at 24 bit you don't add dither.
If you edit a 24 bit wave file and resave it at 16 bit you could add dither before saving at 16 bit.

There's a controversy about the point in adding dither at all, and I can't hear the difference. I can always normalize silence with dither and hear what it sounds like, but not in the context of the music
(JOKE)But then again I don't have golden plugs on my cables.
You obiously can hear it with great cables, a little bit more expensive than what you can afford will do. So if you have 300 dollars you will get away with 400 dollar cables, that's the basic rule (/JOKE)

If you edit your 16 bit file that has been dithered you don't add dither again.
As most audio editors are 32 bit floating point you can always do what you please with the file in terms of effects and edits. It's just like editing any other 16 bit file.
It's not considered good practise to convert from 24 bit to 16 bit and edit from there, but I doubt you will notice much (if any) difference.
What could be noticed is if you add dither several times to the same file, but this depends on the dithering method I guess. I have tried adding dither to a file in sound forge a number of times in a row. I added dither at least 50 times before the noise exceeded -60 dB. I could hear it in a not so busy part of the song, but I don't make ambient music and don't plan to add dither more than 49 times to the same file. <----(I forgot the joke tag here)

[BehindEnemyDeadlines]

Thanks guys. I'm no super hearing billion dollar cable audiophyle. To me my tracks sound fine but I just wanted to check they actually were and learn a bit more about dither in the process, since I know next to nothing about floating point files beyond that they're better. thanks again.

[IIRs]

I just said what do you do if you start and finish at the same word length, which was meant to include hypathetical examples where, for some reason, someone had been forced to start and finish at 16-bit. Is there any way to mitigate noise caused by rounding. Also, I was curious to know what the result in the loss of precision changing from floating point to fixed would be, whether that also introduced noise and what, if anything could or should be done to mitigate it.

Its not really possible to start and finish at the same wordlength: any processing you do to your 16 bit audio file will occur at much greater resolution (probably 32 or 64bit float) so the final result will have a much greater resolution than the 16 bit file you started with. If you then drop it to 16 bit again you will add an extra layer of quantisation distortion / dither noise.

The best way to mitigate noise by rounding is not to do it! The second best way is to add dither first. The accepted best practise is therefore to use higher bit depths for all but the very last (unavoidable) drop to 16 bit to go onto CD, and to add dither once at that final step.

[IIRs]

If you edit a 16 bit wave file and resave it at 16 bit you don't add dither.

This is not always the case. If your edit consists of just trimming the start and end, then yes you don't need to re-dither. But if you add a fade-in or fade-out / do a cross-fade edit / change the volume level in any way then you will end up with 32 bits of resolution or more. If you need to re-save this as a 16 bit file you should definitely add dither.

[Breeze]

I just like to know how these things work even if, in the real world, it doesn't matter because no-one can hear the difference. I'm still curious to know what process or workflow does the least harm or cures the most. Sorry if these questions are agravating.

From the way this discussion has gone, I believe you don't understand what dithering does or exactly how it works. Do a little research in how it applies to graphics compression and it should become clearer (pun not intended). It's primarily a technique used to preserve low-level information that is typically lost in bit reduction, and was much more valuable when reducing to 8-bit files than it is today typically reducing to 16-bit files.

At 24-bits, your dynamic range is 144 dB. Take an audio file that plays to FS, reduce it by 144 dB and tell me if you hear something...

[cosmicdawn]

Let's say I have a kickdrum sample in 16 bit that I decide is too long for my song. I open it in sound forge, add a fade out at the end, save it to a new file and replace it in my project. You think I should add dither to this file?
I would guess this is something Sound Forge should do for me, as it in that case is SF that creates the need for dithering when it's the internal bit depth convertion that makes dithering necessary? Are you sure that SF doesn't do anything internally about this? This would mean that you should add dithering on any channel in your daw that has volume automations or any processing at all. Even lowering the volume a tiny bit, which you always do.
I really don't buy that.
You would end up avoiding adjusting anything at all in your mix just to get that extra quality that no one will hear, and your mix being all out of balance, which everyone will hear

But in the end, even if someone said it is needed I would never complicate my projects by adding dither on all channels etc.
I even don't use dithering when converting from 24 bit to 16 bit on my masters, but that may be just me.

[C00kie]

Let's say I have a kickdrum sample in 16 bit that I decide is too long for my song. I open it in sound forge, add a fade out at the end, save it to a new file and replace it in my project. You think I should add dither to this file?

The kick drum won't sit in your project mixed in at 0dB. Probably it's channel fader is at -9dB or so. That means that the dithering noise you would add, would contribute a noise of about -105dB to mask the truncating artifacts that are of the same magnitude. This is neglectible in the end product, so I'd say there's no practical need to apply dithering here.

But the final criterium should be: can you actually notice the difference? If not, the question is only academic.

This would mean that you should add dithering on any channel in your daw that has volume automations or any processing at all. Even lowering the volume a tiny bit, which you always do.
I really don't buy that.

And you're right. The source file can be 16bits, but before anything is done to it by the host it's converted to 32bits (by padding it with zeroes) so nothing is really lost when the volume is changed. The intermediary data is in 32bits, so no dithering is required.