Should dithering be applied when creating a fade..

[Imagine]

I'm recording at 24 bit/44.1. Should dithering be applied when creating a fade at the end of a region? It gives you that option when the window pops up.

Thanks

[Imagine]

bump

[poncival]

I did a test once to help me understand this better- at 44.1k, 24 bit, I made some silence by consolidating regions with no regions selected (just dead air was selected) Then I cut the silence in two and crossfaded them, using dither. I turned up the speakers really loud and sure enough there was some dither noise during the fade- it got louder and then quieter as the fade went by. At 44.1k, 16 bit, nothing seemed to happen, any noise or lack thereof was the same as the fade played through. I tried this again on the HD system and I don't seem to hear any noise at 96k, 24 bit, so I use dither all the time now. Sometimes on long fades it seems like a smoother fadeout with dither than without, and if you are fading out to silence without dither you will notice that eventually the sound kind of crackles out and with dither it seems to fade into nothing more smoothly. Cross fading full volume tracks at 44.1/24 bit always seemed to sound less noticible to me without dither though. I couldn't tell a difference at 16 bit. Anyway, there's my two cents.

[Mt.Everest]

I'm recording at 24 bit/44.1. Should dithering be applied when creating a fade at the end of a region? It gives you that option when the window pops up.

Thanks

What has always been unclear to me is the idea that dither is meant for when we are going from one bit depth to a lower one. When you are doing fades in PT, you are only changing the gain along a curve, but the bit depth is not changing, so why does one need to apply dither at all in this case?

MT

[ricupad]

What has always been unclear to me is the idea that dither is meant for when we are going from one bit depth to a lower one. When you are doing fades in PT, you are only changing the gain along a curve, but the bit depth is not changing, so why does one need to apply dither at all in this case?

MT

[/QUOTE]

This interests me, too. But what does 'only changing gain' mean in a computer world? Bob Katz's article explains some of it ( http://www.digido.com/portal/pmodule...er_page_id=27/ ), but could somebody please write a 'Dither and Digital Processing for Dummies'-version for us mere mortals to read & learn... I guess this is why you can choose to use dither with AS-plugs?

R.

[Monte McGuire]

What has always been unclear to me is the idea that dither is meant for when we are going from one bit depth to a lower one.

Exactly right...

When you are doing fades in PT, you are only changing the gain along a curve, but the bit depth is not changing, so why does one need to apply dither at all in this case?

MT

How does one change gain? By doing a multiplication. In the arbitrary case, lets say you have a 24 bit value (your audio signal) multiplied by another 24 bit value (the gain constant). When you take the product, you can get up to a 48 bit word. This is your extra resolution, and this wide internal audio representation should always be dithered before it is truncated back to 24 bit.

Most DSP operations involve multiplication, so thus, most DSP operations should be dithered. The only ones that do not require dither are very simple ones: left shift, copying, and addition without any gain scaling of the sum. That's the entire list - anything else, gain, EQ, compression, whatever, all require dither.


Regards,

-monte-

[Mt.Everest]

How does one change gain? By doing a multiplication. In the arbitrary case, lets say you have a 24 bit value (your audio signal) multiplied by another 24 bit value (the gain constant). When you take the product, you can get up to a 48 bit word. This is your extra resolution, and this wide internal audio representation should always be dithered before it is truncated back to 24 bit.

So you are saying that when Protools adds the extra 24bits (or so, I forget the exact #) onto your 24bit audio file, to be worked with in its high precision mixer, and you choose to do a fade, a NEW audio file (24bit) is being created to represent the fade. So dither should be introduced before the truncation of the once 48bit word in the mixer down to the new 24bit audio file which has become your fade...

Am I right (loosely, I know) on this?
thanks
MT

[losttimestudios]

I've always wondered about this too. We have always heard don't apply dither your last step. So assuming the fade is in a musical piece that will be bounced to 16 bit the stereo track and fuklky dithered at that point, shouldn't one wait until then to apply any dithering?

[Monte McGuire]

I've always wondered about this too. We have always heard don't apply dither your last step.

The old wives tale is "don't apply dither _until_ your last step". It's not correct either.

So assuming the fade is in a musical piece that will be bounced to 16 bit the stereo track and fuklky dithered at that point, shouldn't one wait until then to apply any dithering?

You need to dither before each and every truncation from a wide resolution to a low resolution. A typical production path using many plugins and rendered audio tracks (e.g. fades) might have this happen dozens of times. If you don't dither a truncation, you get nonlinearities. If you do add dither before the truncation, you convert what could potentially be distortion into uncorrelated noise. If you're worried that this extra noise from dither might be significant, then the nonlinearities will also probably be significant and are worth avoiding. If the nonlinearities wouldn't have been important to avoid, then similarly, the added noise will also not be important, making the cost of dither negligible.

I see no reason to avoid dithering every truncation, except that sometimes vendors don't provide tools to do this properly, so you have to live with the potential nonlinearities. Fortunately, this is changing, as is how people understand this issue.

What the old wives tale should be thought of as saying is that it's good to avoid multiple wide to narrow resolution steps, and to make the narrowest steps as wide as possible. You can easily do this now by using 24 bit audio files and paths between processing stages. What's not kosher is to think that any one required dither step can somehow be avoided because others have been done. It doesn't work that way. Once dither has been used and the result truncated, you can't expect that signal to "self dither". It is simply not random enough to avoid nonlinearities. Also, if it does happen to be random enough (e.g. 8 bits worth of noise at the bottom of the signal), then surely proper dither will be absolutely microscopic in size and completely benign.

My summary: dither before every truncation. It's a simple rule, easy to remember and apply, and it can't be avoided if you expect the highest quality results. You can sometimes get away with not doing this at every stage, but there's little benefit to doing so.


Regards,

-monte-