Dither Question

[Nibiru]

Bueller...?

[buzzsmith]

Great thread.


Sent from my iPhone using Tapatalk

[FMiguelez]

From what I gather from the Magic Dave post, is that no noise shaping happens when converting from the Soundbites window, only truncating. Noise shaping happens when bouncing to disk, if you so chose to have that option selected within your limiter plugin. Correct?

Hmmmm... I don't know.
Be careful with the terms. Noise Shaping is a property of dither. You can choose to use dither with or without noise shaping. I understand NS is not always desirable anymore. I heard more things against using it than in its favor.

So basically, you either, dither and live with a tiny amount of added noise to mask (and even prevent) quantizing distortion, OR you choose to live with the quantizing distortion (which in 24 bits would be way down there) if you don't dither.
There's no other alternative when reducing bit depths.

Either way, the noise or the distortion are extremely low, and most people would be hard pressed to tell them apart when AB-ing.

My question is...what does one do when recording via sends on an Aux (2bus) to a stereo track, live? My projects are at 32bfp/44.1. Would setting the limiter plugin at 24bit, recording said song, then converting from 32 to 24, be the same as bouncing to disk? I've never been a BTD guy since DP6 and haven't looked back. Am I missing something?

That's an interesting one.
I'm not sure what happens in this case.

Let's examine this...
DP will record, merge and freeze audio at whatever setting you have it for the specific current project (32 bFP, 24 or 16 bits). This setting is INDEPENDENT of DP's native resolution, which is always 32 bits floating point, even if you import 8 bit files.

So if you record in real time, from a send in your case, DP will record at whatever setting you have it as mentioned above, with the actual resolution of whatever you set it in the limiter.

So, and I'm just thinking out loud and speculating here, if your DP session is set to 32 bits FP, and your last insert in the recording track is something like Waves' L2, and you set it to quantise to 24 bits (with or without dither), what you would obtain is a recorded 32 bit FP file with the actual resolution of a (dithered or undithered) 24 bit audio file (because of DP's and L2's settings).
I suppose it would be a 32 bFP representation of a 24 bit file, so this would be a waste of space and would be VERY confusing when managing your final mixes and masters.
If this is so, I would call doing this a definite NO-NO!

What I would do, if my reasoning above is correct, is to match DP's setting to whatever your quantisation resolution in the L2 is.
So if you want a TRUE and real 24 bit file, I would switch DP from 32 bFP to 24 bits (in your case) before recording in realtime like this, so then you really get a 24 bit file.
If you re-import this file into your session, and you do anything to it, DP will always process it at 32 bFP.

Also, this is exactly the reason I chose to print my stems and final mixes at 32 bFP, to avoid unnecessary intermediate noise or distortion stages (by dithering or truncating/quantizing, respectively), and to leave it at its fullest resolution until the final mastering stage.

In my case, when I want to print my stems and final mix, all I have to do is switch DP's setting from 24 bits to 32 bFP (because I choose to print my regular audio tracks at 24 bits, not 32), and record them away. Afterwards I switch back to 24 bits.

What I would REALLY like to know, and so far nobody has given us a definitive answer, is:
What happens if you record in real time using 24 or 16 bits in terms of dithering? Does DP automatically dither from 32 bFP to 24 or 16 bits as it prints it, or does it TRUNCATE it and leave the quantisation distortion?
I mean this in cases when printing or recording in realtime groups of previously recorded tracks, like submixes, stems, final mixes, etc., to save CPU and simplify things, for instance.

I certainly hope (and expect it to be) the former...

Would the dithering command in the menu be responsible for this?

I REALLY want to know this...

[Killahurts]

I admit I didn't read the whole thread word for word.

I just wanted to add that the dither option in the audio menu is only applied when there is DP processing going on.

In the old days, there wasn't anything for me but Apogee's UV22 dither. Now there are many, but when you look at what's really happening, you have to make a judgement on whether or nor to use dither. The "noise shaping" is merely the EQ of the dither, more or less- high or low frequency gain. If the dither is mid-range heavy, you might hear it, hence the noise shaping to get it to the less perceived frequencies. It is still added noise to lessen distortion between the lower bits. If you are doing music that constantly occupies the higher bits (for instance, hard rock or EDM), it doesn't make much sense to add dither/noise. It is when there are finer details that you don't want to get what I used to describe as "crunchy" audio. Reverb tails are the worst, because the tail can fall off the audio "cliff" and suffer the fate of Wile E. Coyote.

I always dither before going to a 16 bit format, because I have been trained that way, I don't take chances. 32 bit to 24 bit? Not so much.. Sorry, no one can hear it.

[FMiguelez]

I admit I didn't read the whole thread word for word.

I just wanted to add that the dither option in the audio menu is only applied when there is DP processing going on.

Yup. MagicDave's writings on that were profusely cited

I always dither before going to a 16 bit format, because I have been trained that way, I don't take chances. 32 bit to 24 bit? Not so much.. Sorry, no one can hear it.

The thing is that you don't necessarily hear the dither noise as noise, but its absence in favour of quantising distortion does collapse the stereo image a tiny bit (or so say some engineers that know more than I do and who have tested these things). Personally I'm not entirely sure about this, since sometimes I think I can hear it and some times I don't.

The advantage of dithering is that the resulting file, if done with something like L2, will give the appearance of having a higher resolution than that with which it was bounced in such a way that a dithered and quantised 16 bit mix will sound more like a 20 bit audio file in terms of bit depth. A dithered 24 bit audio file would be more like 28 bits. This according to Waves' manual.
Also, being technically correct is not a bad thing regarding this particular issue.

Dithering, or dealing with it properly to totally avoid it (by printing important things at 32 bFP), is extremely easy and does not cost one anything except for a minimum of effort.

Based on all I've read about it, I think it is totally worth it to dither from 32 bFP to 24 bits, at least for the final mix and the stems.

But the best thing is to keep it at 32 bFP as long as possible.
Why introduce unnecessary noise or distortion during intermediate stages? I'm all for keeping the highest resolution until the mastering end!

[Shooshie]

What I would REALLY like to know, and so far nobody has given us a definitive answer, is:
What happens if you record in real time using 24 or 16 bits in terms of dithering? Does DP automatically dither from 32 bFP to 24 or 16 bits as it prints it, or does it TRUNCATE it and leave the quantisation distortion?

I certainly hope (and expect it to be) the former...

I'm the last one to claim expertise on this, but it's my belief that you don't dither until you print. That is, any file that leaves the DP domain to join the free world must be dithered.

When you record, a file is printed to disk. DP isn't converting it from 32 bits to 24 bits, it's merely recording it at 24 bits. Later, DP handles that file at 32 bits fp when combining it with other files in mixing, or when doing operations on audio such as adding plugins, fading, panning, and so forth. Dithering isn't part of the equation at the mixing stage. Each file interacts with the other files at its full recorded resolution.

Dithering is like spraying unwanted grid-lines in a picture with fine spray paint it reduces the contrast of those grids until they don't really matter. It's of such high frequency that we don't hear it, but that's not what matters. Those unwanted lines in audio are waves, and they are like a grid. Overlap two grids and you get a Moire pattern. You may not see the faint lines of the grids, but you can darn well see the moire (aliasing). And that's what we're eliminating when we dither those ultra-sonic waves with our digital spray paint. (noise) Eliminate the grids, and you eliminate the moire.

Unlike the picture, an audio file does not get gummed up with gray when you spray paint those tiny particles of noise. The ultra-sonic white noise fits comfortably around the larger waves, moving with it and becoming something like a canvas in front of speakers, vibrating with the sound. As long as it does this above our hearing range, and as long as it doesn't adopt a regular pattern (another grid, another moire), then it's irrelevant to what we hear.

But we don't want to gum up those higher frequencies until we're done with them. As long as they are harmonic, the "aliasing" they create is also harmonic and timbral. So, when we're mixing we allow for a great deal of overhead until we're ready to print a mix to disk.

I don't think of dithering for myself or for my personal listening to the files that I create, because I listen to those files at more or less the same sample and resolution at which i created them. Dithering isn't really necessary. I dither for future applications of anything that leaves my domain. Who knows where it will end up? Who knows what someone may combine with it later on down the road? Or if I'm converting down to 44.1K/16 bits, then we're once again overlaying a finer grid over a more coarse one, once again creating a moire pattern of sound, and that requires dithering.

As long as you remain in the DP domain, you have the overhead to utilize all the sound, so dithering is not necessary. Again, when coming directly from a microphone, DP is receiving a sound file in real time, at 48K/24 bits or whatever, and there is no need to dither that. You'd be smudging your picture before you even look at it. Only when you are combining sound files and creating another file from that mix is it necessary to deal with those little ultra-sonic moire-like grid combinations of waves and filters.

If I've misunderstood your question and answered too simplistically, forgive me for my density. I'm just attempting to explain the way I see things in hopes that it may apply to whatever you do not understand.

Shooshie

[FMiguelez]

What I would REALLY like to know, and so far nobody has given us a definitive answer, is:
What happens if you record in real time using 24 or 16 bits in terms of dithering? Does DP automatically dither from 32 bFP to 24 or 16 bits as it prints it, or does it TRUNCATE it and leave the quantisation distortion?

I certainly hope (and expect it to be) the former...

I'm the last one to claim expertise on this, but it's my belief that you don't dither until you print. That is, any file that leaves the DP domain to join the free world must be dithered.

When you record, a file is printed to disk. DP isn't converting it from 32 bits to 24 bits, it's merely recording it at 24 bits. Later, DP handles that file at 32 bits fp when combining it with other files in mixing, or when doing operations on audio such as adding plugins, fading, panning, and so forth. Dithering isn't part of the equation at the mixing stage. Each file interacts with the other files at its full recorded resolution.
Shooshie

Hi, Shoosh.

I should've been clearer. Sorry about that.

I am exclusively talking about printing submixes, printing stems and final mixes in real time that come from other (previously recorded) audio tracks.
So, if I have a bunch of printed string tracks, and I want to record that stem, DP will be processing all that stuff internally at 32 bFP. But if I print it, someone needs to do the truncating, and it may have or not dither.
That's what I want to know.

So in cases like this, I think dithering has A LOT to do with the mixing stage, at least with the final stages of mixing

I'll edit my post to clarify what I mean.

[Shooshie]

I should've been clearer. Sorry about that.

I am exclusively talking about printing submixes, printing stems and final mixes in real time that come from other (previously recorded) audio tracks.
So, if I have a bunch of printed string tracks, and I want to record that stem, DP will be processing all that stuff internally at 32 bFP. But if I print it, someone needs to do the truncating, and it may have or not dither.
That's what I want to know.

So in cases like this, I think dithering has A LOT to do with the mixing stage, at least with the final stages of mixing

Well, I think what I said still applies. If those strings have left DP, they need to be dithered. If you received strings from Pro Tools, it's up to them to dither their mix up to that point. When it leaves the Pro Tools domain, for the free world, it should be dithered, and you will pick it up and use it in DP.

If, on the other hand, you or someone else creates strings in DP, and if you or someone else then imports them into DP, it really hasn't left the DP domain if you print it at 32 bits FP and the same sample rate. But the deeper I go, the more I see your point. You need a definitive answer to several "what-ifs," and I'm not the guy to do that. That's Magic Dave's domain, or someone who is constantly doing this sort of thing. My projects tend to start and end in DP, up to the point of printing final 2-tracks for the world, at which point I dither (or rather, I tell DP to dither).

Shooshie

[FMiguelez]

I mean, I would be very surprised (not in a good way) if DP didn't automatically dither as it prints audio from other tracks when submixing.

Can you imagine if someone, for any reason, is working at 16 bits? If someone decides to print in real time say, the brass tracks, to free up resources and dealing with only one track instead of 10, I really suppose DP dithers whatever 32 bFP signal is coming out of the busses to the print track to 16 bits instead of simply truncating and leave us with the quantisation distortion

But like you say, that's MagicDave's territory. Or MOTU could tell us everything by simply stating it in the manual and CORRECTING the entries about the Dither command that have caused confusion for YEARS now.

So far, all we can do is assume, suppose, and hope. Or test these things. I think Dan Worley started testing this, and if memory serves me right, he was not happy with his findings...

[stevenew]

Interesting thread, but I am still confused (always easy to do).

Example: I have DP set to 44.1k/24 bit, I record all my audio, I keep all my VI in real time not bounced to audio files. All the recorded audio tracks and real time VI tracks are bussed to their relevant group aux, these in turn are bussed to a final mix down aux (final insert effects added here) then bussed to a stereo audio track so I can print in real time a mix. This 24-bit printed mix is then exported from the Soundbites window in to the real world.

1) When printing this 24-bit mix should the DP dither in the audio menu be turned on?

2) Should I be using a 3rd party (Ozone) dither set to 24bit as my final insert on the mix down aux?