To Dither or Not To Dither, that is the question

[zaratero]

@David, if I understand your post correctly, in a 24 or 16 bit project default scenario, if a new file is created by a merge command, all the truncation from 32 float to 24 bit is taken care of (dithered, because it should be, right?) independently of the dither setting of our choice. Is this correct?

@umbrella, can you find MagicD´s post on the subject about it working under the setting?

If we are right (or wrong, but clear about this), then the question would be more about sonics and less about actual funtionality, which I still think is unclear.

My feeling after this chat is that this setting does determine merges be dithered or not. We are asked if we bounce or export, so what´s left?

IMMO a little more comprehensive chapter in the DP8 manual would be nice.

[Kubi]

Generally, MOTU is pretty clear that they recommend to leave dither ON. DP's dither doesn't apply to operations under MAS (i.e. recording, re-recording) but it does apply to operations under DP itself (merging, pitch shifting, etc.) In all those instances you're truncating, and ditherless truncation noise is always worse than dither noise. That's why we dither in the first place!

In general, there is so much less noise in the audio chain than there used to be in the analog days, and truncation noise is such an ugly artefact, I'd much rather have the occasional double-dither than risk truncation noise.

The question remains open if MOTU's recommendation still holds when you choose to work in 32bit float. But for those of us working in 24bit or 16bit (God forbid! Why would you do that?), leave dither option checked.

The only time I turn DP's dither off is when I export a soundbite from the soundbites window. Not 100% sure, but I believe that is a DP function. When exporting a soundbite unchanged I of course don't want dither added, so I turn it off, to be safe. Would love to hear from MOTU/MagicD if DP's dither (when checked) gets applied when exporting a sound bite at its original bit depth and sample rate.

[Dan Worley]

Generally, MOTU is pretty clear that they recommend to leave dither ON. DP's dither doesn't apply to operations under MAS (i.e. recording, re-recording) but it does apply to operations under DP itself (merging, pitch shifting, etc.) In all those instances you're truncating, and ditherless truncation noise is always worse than dither noise. That's why we dither in the first place!

In general, there is so much less noise in the audio chain than there used to be in the analog days, and truncation noise is such an ugly artefact, I'd much rather have the occasional double-dither than risk truncation noise.

The question remains open if MOTU's recommendation still holds when you choose to work in 32bit float. But for those of us working in 24bit or 16bit (God forbid! Why would you do that?), leave dither option checked.

The only time I turn DP's dither off is when I export a soundbite from the soundbites window. Not 100% sure, but I believe that is a DP function. When exporting a soundbite unchanged I of course don't want dither added, so I turn it off, to be safe. Would love to hear from MOTU/MagicD if DP's dither (when checked) gets applied when exporting a sound bite at its original bit depth and sample rate.

I think you hit the nail directly on the head.

Thanks for that.

c-ya,

Dan Worley

[Umbrella]

There is no need for DP to do any dithering in
its own 32-bit floating point environment.

File "leave dither checked" in the same drawer as "stop ___ or you'll go blind".

What about DP dithering from 32-bit float to 24-bit integer for D/A (in a 24-bit project)? Hopefully DP does this even when the menu option is "off", yes?

In 5.13 fade files will be chopped down to 24-bit without the option checked (as far as I understand) - this wouldn't be a good thing, would it?

I could be wrong, but I feel we really need someone from MOTU to clear this up...

[Dan Worley]

The only time I turn DP's dither off is when I export a soundbite from the soundbites window. Not 100% sure, but I believe that is a DP function. When exporting a soundbite unchanged I of course don't want dither added, so I turn it off, to be safe. Would love to hear from MOTU/MagicD if DP's dither (when checked) gets applied when exporting a sound bite at its original bit depth and sample rate.

Here's something perplexing:

I'm not so sure if Dither is working at all during an export.

Keep in mind I'm using DP 6.03.

I took a 24-bit mix. I exported it twice as 16-bit from the soundbites window, the first with Dither off, the second with Dither on. I imported the resulting files into a 16-bit project. Auto Conversion off. The files were not changed in any way. Did a null test. The two tracks canceled out completely. I have the monitors cranked to 12. There is absolutely no noise. No hiss, no hum, no phantom noise, no clicks, no pops, no rough edges. There is no residual noise at all... but there should be.

So, no matter if it's checked on or off, either Dither is not working at all during export, or it's working all the time.

I did this test three times, all with the same results.

c-ya,

Dan Worley

[David Polich]

Here's the deal -

Leave Dither OFF. Period.

And print everything in 32bit float unless you have to convert it to 24 bit integer.

Hah-hah. You're joking, right?

[mhschmieder]

It might be time to link to Magic Dave's take on this (circa 2008?)...

[Kubi]

The only time I turn DP's dither off is when I export a soundbite from the soundbites window. Not 100% sure, but I believe that is a DP function. When exporting a soundbite unchanged I of course don't want dither added, so I turn it off, to be safe. Would love to hear from MOTU/MagicD if DP's dither (when checked) gets applied when exporting a sound bite at its original bit depth and sample rate.

Here's something perplexing:

I'm not so sure if Dither is working at all during an export.

Keep in mind I'm using DP 6.03.

I took a 24-bit mix. I exported it twice as 16-bit from the soundbites window, the first with Dither off, the second with Dither on. I imported the resulting files into a 16-bit project. Auto Conversion off. The files were not changed in any way. Did a null test. The two tracks canceled out completely. I have the monitors cranked to 12. There is absolutely no noise. No hiss, no hum, no phantom noise, no clicks, no pops, no rough edges. There is no residual noise at all... but there should be.

So, no matter if it's checked on or off, either Dither is not working at all during export, or it's working all the time.

I did this test three times, all with the same results.

c-ya,

Dan Worley

Thanks! It's entirely possible that DP's dither does not get applied for exports. Interesting indeed!

[Kubi]

It might be time to link to Magic Dave's take on this (circa 2008?)...

Here's the quote, this is from an email I believe? Not sure, I copied it into my DP Manual folder, but it is un-edited Magic Dave:

"Dither under the Audio menu is applied when DP creates an audio file. There is a difference between DP creating an audio file or that audio file being created by MAS. For example, recording, rerecording, or bouncing is a MAS function. Pitch shifting, creating crossfades, printing a destructive effect, or merging audio are all DP functions. When DP creates an audio file directly, it is going from the floating point number to either a 16 bit or 24 bit fixed integer sample. Rounding happens automatically. The only difference between the dither in the Audio menu and the dither in the MW Limiter is that the Audio menu dither doesn't have the sub-option of noiseshaping."

[Frodo]

For example, recording, rerecording, or bouncing is a MAS function.

Wow. I must go in search of more info. I honestly never considered to distinguish DP functions from MAS functions. Boy, do I feel ignorant after all these years!!

(I'm not being at all facetious-- I'm genuinely only about 75% sure of the differences.)

[Shooshie]

While I defer to whatever Magic Dave says, my rule of thumb in the past was always to dither when leaving the DP environment. Bouncing to disk, mainly, or exporting audio files. You don't want audio with loose ends -- dangling aliasing. I need to reread Dave's reports on the subject. I remember them being quite enlightening. I think I've got them stored around here, and maybe even posted in the Tips Sheet thread. (one of the later pages, if they're there)

Shooshie

[Tritonemusic]

Here's a longer post from Magic Dave about the subject:

1. Insert the MW limiter in the last plug in slot on your Master fader

In DP 2.7 or earlier, you would also want the fader at unity gain.In DP3, you can change the position of channel inserts to post fader (they were originally pre fader). There's a little handle at the bottom of the channel inserts. If you drag this up, inserts below that handle are now post fader. That means you could change the fader level and you would be adjusting signal going into the limiter, as opposed to changing the output level of the limiter.

2. Make that insert slot Post Fader (DP 3 only)



3. In the MW limiter, choose 16 bit (Quantization)

When you go from a higher bit depth to a lower bit depth, what you don't want to do is just chop off the bottom bits. If you make a 24 bit sample into a 16 bit sample by losing the bottom 8 bits, that's called truncation and you are losing any info in those bottom bits. The better way to do it is to quantize the higher number to a lower number. Essentially you are rounding off the bottom 8 bits into the next 3 or 4 bits above that. In other words, the resulting 16 bits contain a rounded version of the former 24 bit sample and that rounding ended up in the bottom 3 or 4 bits of the 16 bit sample. An unfortunate side effect of bit rounding is distortion of signals that occupy those rounded bits. It's important to understand that quantization only effects the bottom of the sample and therefore only effects very soft signals that are represented by those bottom bits. A loud signal that lives in the top of the sample is not affected by the quantization.

The MW limiter works at 64 bit float. It then rounds down to 32 bit float for output into the MAS signal path. MAS rounds down to 24 bit fixed integer for output to D/A converter and print to file.

However, notice the little button in the bottom left hand corner of the MW Limiter? Bit quantization is always on in this plug. The maximum bit quantization is 24 bits. That means that even though the sample outputting from the MW limiter is 32 bits, the sample has been quantized into the top 24 bits. The whole point of the limiter is to do exactly that, limit the top end of the signal. The MW limiter does not add any gain. It can't. And what it will do is limit any input signal to 24 fixed bits.

When rounding from floating point to fixed integer, dither is not required.

I always have a Limiter on a master fader when I work in DP. I usually set the limiter to 24 bit while I'm tracking, then drop to 16 bit as I start to mix. If you just want to forget about it, leave it at 16 bit.

If you are not doing your own mastering, it is very much desirable to provide a 24 bit file to the mastering house. Obviously for that you would leave the Limiter at 24 bit and BTD at 24 bit.



4. Dither is optional. Noiseshaping is a sub option of Dither

Dither is the process of randomizing the bottom bits. This randomization produces noise. Dither is noise. This process obscures quantization distortion. If you had a signal that was suffering from quantization distortion, applying dither would soften the distortion but introduce low level noise to the audio.



Noiseshaping is a variation of the randomization and the result is a different spectral shape of the dither noise. Think of it as the difference between white and pink noise. There's no mystery as to what dither noise sounds like. Set the quantization of the MW Limiter to 4 or 6 bits. Now turn on dither. You will hear the dither noise. Select noise shaping. Now you hear the difference in the frequency curve of the dither noise. When you go from a higher bit depth to a lower bit depth, you should always quantize. Dithering however, should be considered an option, not a requirement. Dithering should be applied if the program material benefits from it. Noiseshaping should be selected based on what you hear (and prefer). One example of when I would not use dither is highly compressed rock music. If the tracks are just loud all the way through, you may never be using the bottom bits of the sample. Adding dither would only introduce noise and would not fix any distortion.

The time to be picky about dither and noiseshaping is when you have an audio track with a wide dynamic range. The best example to illustrate this is a single piano note. As the note decays, it uses lower and lower bits in the sample. When the note gets into the bottom 3 bits of a quantized 16 bit sample, you'll hear the quantization distortion. If you add dither, you won't get the distortion but you will hear low level noise (kinda like tape hiss). You make the choice as to the noise shape curve.

If you care about these subjects you should do basic tests like recording a piano note so you actually hear what these processes do. Just following rules about how to engineer won't get you good sounds. The experience of actually hearing the difference gives you the expertise to make an educated decision about what processes you should use.



5. Choose 16 bits from the Bounce to disk window

Choosing 16 bit in BTD means that DP writes a 16 bit file. Since the output of DP is always 24 bit, a 16 bit BTD just truncates the last 8 bits. You want to make sure there is no useful information in those last 8 bits before you BTD. That's where the MW Limiter comes in. The output of the master fader to D/A converter or file is always 24 bit. If the Limiter is used and set to quantize to 16 bits, the output to file or D/A converter is still 24 bit, but the last 8 bits are now zeroed out. They no longer contain audio data. What had been 24 bits of info has now been rounded into the top 16 bits. That means that when BTD truncates the bottom 8 bits, it's OK because there's no data in those bits.

When you choose the output for a bounce, DP is rounding to 24 bit fixed. If the bounce is to 24 bit, you get all 24 bits. If the bounce is to 16 bit, the bottom 8 bits are truncated by the bounce process. That's why you should use the MW limiter to quantize to 16 bits, even if your source files are 16 bits.



6. Dither option under the Audio menu does nothing when you BTD

When you bounce to disk, the Dither option under the Audio menu does nothing.

There's been lots of talk about what the Dither option under the Audio menu does. I asked The Person Who Knows and got this reply:

Dither under the Audio menu is applied when DP creates an audio file. There is a difference between DP creating an audio file or that audio file being created by MAS. For example, recording, rerecording, or bouncing is a MAS function. Pitch shifting, creating crossfades, printing a destructive effect, or merging audio are all DP functions. When DP creates an audio file directly, it is going from the floating point number to either a 16 bit or 24 bit fixed integer sample. Rounding happens automatically. The only difference between the dither in the Audio menu and the dither in the MW Limiter is that the Audio menu dither doesn't have the sub-option of noiseshaping.



-Magic Dave

[Shooshie]

The earliest link I can find to that is:

http://www.motunation.com/forum/viewtop ... 028#p51028" onclick="window.open(this.href);return false;

It's strange that we can no longer access the original post. I guess it is one of those bits that was lost to one or another crash of the old Unicornation many years ago. The date of the above post was 2005, and the poster who quoted Dave said it was from an old post. Anyway, it's interesting information.

It's always seemed to me that DP internally is very, very powerful -- too powerful to interface with hardware or conventional audio files. MAS, on the other hand, is made for interfacing with hardware, converting audio, running the plugin layer, and utilitarian tasks like time stretching. I don't know where the line is drawn between them, and I suspect there is a lot of overlap in what they can do as well as which does what in different situations, so it may be futile to try to own this information. I'd take Dave's word as the gospel, though.

Shooshie

[Dan Worley]

So, no matter if it's checked on or off, either Dither is not working at all during export, or it's working all the time.

I did this test three times, all with the same results.

c-ya,

Dan Worley

Thanks! It's entirely possible that DP's dither does not get applied for exports. Interesting indeed!

Yeah, I'm pretty sure it's not, and all this time I thought it was. At least we won't have to worry about shutting it off before an export from now on, so it's good to know.

c-ya,

Dan Worley

[Dan Worley]

It might be time to link to Magic Dave's take on this (circa 2008?)...

Here's the quote, this is from an email I believe? Not sure, I copied it into my DP Manual folder, but it is un-edited Magic Dave:

"Dither under the Audio menu is applied when DP creates an audio file. There is a difference between DP creating an audio file or that audio file being created by MAS. For example, recording, rerecording, or bouncing is a MAS function. Pitch shifting, creating crossfades, printing a destructive effect, or merging audio are all DP functions. When DP creates an audio file directly, it is going from the floating point number to either a 16 bit or 24 bit fixed integer sample. Rounding happens automatically. The only difference between the dither in the Audio menu and the dither in the MW Limiter is that the Audio menu dither doesn't have the sub-option of noiseshaping."

Oh, good. I was looking for that but couldn't find it. Thanks!

c-ya,

Dan Worley