The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Good dither practices, what are yours? Dynamics Plugins
Old 8th January 2017
  #31
You convert the wordlength within tracks of a mix?! How's that possible? Edit: Oh, analogue inserts maybe?

AFAIK, this only happens when exporting to a fixed point format or sending out audio to the convertors (always fixed point). It never happens inside the DAW, the latter being typically floating point (32bit fp or 64bit fp) and generously sized. Oh, and floating point can't be dithered anyway!
Old 8th January 2017
  #32
Gear Maniac
 

Quote:
Originally Posted by FabienTDR View Post
You convert the wordlength within tracks of a mix?! How's that possible? Edit: Oh, analogue inserts maybe?

AFAIK, this only happens when exporting to a fixed point format or sending out audio to the convertors (always fixed point). It never happens inside the DAW, the latter being typically floating point (32bit fp or 64bit fp) and generously sized. Oh, and floating point can't be dithered anyway!
Yes, hardware inserts or analogue monitoring, so you have to put a dither plugin before them. Doesn't seem like most DAWs do this automatically.

What I'm not sure about is recording output of VST instruments in a 24 bit project, for example.
Old 8th January 2017
  #33
Lives for gear
 
grrrayson's Avatar
 

Quote:
Originally Posted by Plush View Post
They should be using dither then
How does that look for you in practice, though—how do you personally actually do that? It would seem the way to accomplish that is to have a dither plug-in live in the last insert of your stereo buss.
Old 8th January 2017
  #34
Gear Nut
Quote:
Originally Posted by Plush View Post
Gain changes, bounces, renderings, using plug ins, eq's, ---all require dither to 24 bits again.

r.
But that means that i need a dither plugin between all my plugins in the inserts or i can put just one at the end of the insert chain?
Also where the dither plugin go, after or before the plugin fx?
Putting one dither plugin in the master/mixbuss it isn't the same thing?
I am very confused, always thought that the only time i needed dither it was before the final mixdown.
Btw, i dont bounce to track (still in PT 11, but i know in 12 its safe to do it) i record all my tracks into a new one, so in that scenario, where the dither plugin go?
Old 8th January 2017
  #35
Lives for gear
 
JulenJVM's Avatar
Quote:
Originally Posted by Aumumano View Post
But that means that i need a dither plugin between all my plugins in the inserts or i can put just one at the end of the insert chain?
Also where the dither plugin go, after or before the plugin fx?
Putting one dither plugin in the master/mixbuss it isn't the same thing?
I am very confused, always thought that the only time i needed dither it was before the final mixdown.
Btw, i dont bounce to track (still in PT 11, but i know in 12 its safe to do it) i record all my tracks into a new one, so in that scenario, where the dither plugin go?
One instance on the Stereo out should do, others put it on the monitoring path, depending on what you want to do.
Old 8th January 2017
  #36
Quote:
Originally Posted by Aumumano View Post
after or before the plugin fx?
You don't. You do it only when truncating fixed point audio. Only then. This is only the case when explicitly exporting your audio to a fixed point format, or sending audio to hardware devices (CD, DA, etc). If ever, and your DAW doesn't export properly on its own, and you are not exporting to a floating point format, put a dither into the last slot of the master bus.

Don't dither within a mix (except hardware outputs for inserts of course). That's literally silly. It reduces your signal to noise ratio, without any advantage.

All modern DAWs and plugins use a standard floating point format, which is radically different to fixed point. You can't dither floating point. Truncation effects are generally rather irrelevant in this world, it's hardly measurable.

Last edited by FabienTDR; 8th January 2017 at 11:23 PM..
Old 8th January 2017
  #37
Gear Guru
 
Karloff70's Avatar
 

So, this is starting to simplify.

They way I hear it now is, do two things. Put a dither plug last in the chain on the master before it goes out, and put one just before any hardware inserts. Done.

Right?
Old 8th January 2017
  #38
IMHO, yes.

Dither before the hardware outputs (hardware outputs, track export to fixed point, redbook CD creation, etc). We can generalize it this way because we can assume that all digital hardware outs are fixed point at the end of the day.

Dither all fixed point data before it gets processed (e.g. gain changes, limiting). These cases are unusual in practice, and for the most cases, straight impractical. But definitely worth the mention, as some old DAWs worked this way (often with tricks and compromises). Don't do this in floating point DAWs!

Dither all fixed point data truncations. E.g. when converting 24bit fixed point to 16bit fixed point.
Old 9th January 2017
  #39
Lives for gear
 
grrrayson's Avatar
 

Firstly, Fabien, I really appreciate all your intelligent and bull****-free input on this forum. Moreover, Kotelnikov is my desert island compressor—I love what you do. Really brilliant.

Quote:
Originally Posted by FabienTDR View Post
Dither all fixed point data before it gets processed (e.g. gain changes, limiting).
Respectfully, this seems misleading, as it implies a dither plug-in should always be first in the chain. When you record in a 24-bit session, the recorded file is 24 bits, as the converter is working at 24-bits, so according to what you just said any basic recording with any fader adjustments would be a case for having a dither plug-in first.

A better rule of thumb is what you said above: always dither as a last stage when converting to a fixed point format, whether from floating point (e.g. to a fixed-point DAC for monitoring or analog inserts), or from one fixed point format to a lower bit depth fixed point format.

Right?


Also, to everyone: where is a definitive list of which DAWs dither properly (and how do we even know whether they do or not)? It seems we should make a running one that can be updated over time.
Old 9th January 2017
  #40
Ah! I already feared someone would quote me without the following "These cases are unusual in practice, and for the most cases, straight impractical."

Fair enough, indeed, doing this right can quickly turn into an complicated mess. The 24bit AD / DA scenario complicates it even further of course. The key point was more about the fact that most professional audio applications do all their processing in floating point, because it is far less sensitive to truncation (in typical audio workflows).

Thanks for the flowers! That's highly appreciated.
Old 9th January 2017
  #41
Lives for gear
 
grrrayson's Avatar
 

Quote:
Originally Posted by FabienTDR View Post
indeed
Right.

Although it's a tangent, I'm curious now, and hearing the explanation might avert confusion on the part of people superficially skimming the thread: why would you ever need to dither first while staying at the same bit depth?
Old 9th January 2017
  #42
"why would you ever need to dither first while staying at the same bit depth?"

Given a fixed point environment (which is rather difficult to find in 2017), it can make sense. I really just mentioned it for academic purposes.

Reducing gain (in a fixed point environment) is equivalent to a fixed point truncation. A such, it has to be handled in the same way.



Say, you have a 16bit system and want to reduce gain by ~12dB. You divide each sample by 4, effectively trying to extend the wordlength to 18bit. The last two bits don't fit, so they get truncated (ignored) during the process. The fact that the lowest bounds now have been cut away means we'll now likely miss "a step of the ladder" in a very regular and predictable manner.

This error (also called quantization distortion) has a direct dependency on word-length, sample-rate, signal level and frequency. This is a nonlinear dependency capable of adding tones that weren't in the original material (as all nonlinearities do), and generally unwanted. At least for non-creative applications.

It's the regularity of the "errors" that makes them so problematic. Truncation without dithering is a nonlinear system with very annoying properties.

This nonlinearity, including all its drawbacks typical of nonlinear systems, can be fully neutralized by adding a properly scaled, fully random (!) signal to the original signal. Upfront. This works because it now reliably "shakes" the whole system in such a manner that no regular patterns can appear. Ever.

Baam! We now have a linear system with noise. Linear systems have very favorable properties, especially in audio scenarios. That's a priceless advantage.

Ok, I explained dithering along the way, but you get the point. Truncation and reducing gain is equivalent.

Things get immensely complicated as soon dynamics processing or recursive algorithms (most filters) come into play. That's what I meant above with "impractical".




On the subjective side, dithers have the potential to affect tone/timbre and relax low level stereo correlation. This explains why proper dithering, i.e. the lack of quantization distortion is not always experienced as superior. It also explains the range of effects it can produce, it is in fact very similar to jitter errors and reviews of clocks often share the same language as dither talks.

In any way, a properly scaled TPDF dither will have the advantage to fully preserve the original tonality and stereo image in any given case. That's my personal preference. Not because I listened and preferred it. But because it is the only solution that frees my mind for the more relevant questions!! :D

Last edited by FabienTDR; 9th January 2017 at 02:05 AM.. Reason: Some simplifications
Old 9th January 2017
  #43
Lives for gear
 
Plush's Avatar
Quote:
Originally Posted by grrrayson View Post
How does that look for you in practice, though—how do you personally actually do that? It would seem the way to accomplish that is to have a dither plug-in live in the last insert of your stereo buss.
I use the SADiE editor and in that daw I can assign a global dither (24 bit) to be used when processing with level changes, eq, dynamic range control, or in a bounce, a render, or with a plugin. Typically dithering 24 bit to 24 bit. I can select the type of dither I want but typically I use triangular dither.

It sits last on buss 1 and 2 and affects the mix when processing is passed through the mixer.

When outputting to cd, I choose a different dither---usually POWR-3 or a Prism SNS variant.
If you are outputting your 24 bit signal to a d/a converter with no level change, there is no need to have dither on the output. (You already dithered it.)
Old 9th January 2017
  #44
Lives for gear
 
JayTee4303's Avatar
Quote:
Originally Posted by Plush View Post
I use the SADiE editor and in that daw I can assign a global dither (24 bit) to be used when processing with level changes, eq, dynamic range control, or in a bounce, a render, or with a plugin. Typically dithering 24 bit to 24 bit. I can select the type of dither I want but typically I use triangular dither.

It sits last on buss 1 and 2 and affects the mix when processing is passed through the mixer.

When outputting to cd, I choose a different dither---usually POWR-3 or a Prism SNS variant.
If you are outputting your 24 bit signal to a d/a converter with no level change, there is no need to have dither on the output. (You already dithered it.)
U87ai>AvalonM5>Input 1 on Motu HD-192. Straight into DAW, AND...

Back out Output 3 of HD-192> UA1176>LA2a>Input 5 HD-192

Straight into DAW from Input 5, AND...

Back out HD-192>Artist's headphone mixer.

Prefer the option of recording both straight, and slightly compressed audio, each take. Could do it all OTB, on the way in, but my time machine is down for repairs.

Since no dither is possible in the HD-192 DSP software, can I assume the move is to control 1176 input with the Avalon M5?

Or am I missing something, and using gain or attenuation in the HD-192 DSP software, between the M5 output and 1176 input won't contribute truncation distortion?

If not why not? (Part of a larger set of questions, where one answer applies to all.)
Old 9th January 2017
  #45
Lives for gear
 

The debate / confusion with dither comes with using dither on hardware inserts for the DAW app. Another debate comes from even needing to run the project at 32 bit float instead of just fixed 24 bit. If you are good with gain staging then you don't need 32 bit float though some debate plug performance (a 32 bit float project setting is basically just a safety net for gain staging). Thus if your converters are set to the same as your project sample rate then there is absolutely no need to add dither at that stage (running 24 bit 48k DAW project sample rate and converters sample rate here out to analog desk for mix / outboard). Dither MIGHT be used for the last stage of lowering the bit depth rate for a final delivery format, i.e. down to 16 bit IF staying ITB. Then the real debate starts as dither is basically adding quasi random low level noise. If you are outputing your 2 buss to outboard for processing another option for skiping dither all together is using a separate 2 channel A to D converter set to the lower bit rate and record that as the mix deliverable or if you can just set the return A to D converter to the lower bit rate when using the same converters / computer.

As there is no real single universal standard (even amoung Mastering) with how people setup their DAWs and converters, individual configurations are going to be all over the map. The best one can do is experiment (hopefully blind with a friend) to see what sounds best given what they use.
Old 9th January 2017
  #46
Lives for gear
 
toledo3's Avatar
 

Quote:
Originally Posted by Aumumano View Post
I am very confused, always thought that the only time i needed dither it was before the final mixdown.
Btw, i dont bounce to track (still in PT 11, but i know in 12 its safe to do it) i record all my tracks into a new one, so in that scenario, where the dither plugin go?
Yes, just mix down in that case.
Old 9th January 2017
  #47
Gear Guru
 
Karloff70's Avatar
 

So it turns out I hear a difference when putting a dither plug before hardware inserts. And I like it. Well worth it, at least with LogicX.

New habit formed. And a still rolling thread to take more new (for me) wisdom from. Nice. Thanks, chaps!!
Old 9th January 2017
  #48
Deleted c1b33bf
Guest
Been following the other threads including the Naturalize dither thread. Very interesting indeed. Apparently Bob Katz has also addressed these issues in his book 'Mastering Audio.'

I've been converting and exporting 24 bit to 16 bit files using Audacity to build libararies on my MPC.

According to this I should be ok as the default had shaped dithering turned on. Thank god.

"Dither applied on Export - detailed explanation

Dither is only applied when converting from a higher bit depth to a lower bit depth. Exporting to WAV or other uncompressed audio format is virtually lossless in all situations. However there may be a very small amount of change to the audio data with some types of uncompressed audio export, depending on the settings in Audacity Preferences (see below).

Exporting to 32-bit PCM WAV format is 100% lossless - no dithering is applied.
Exporting to 24-bit PCM WAV format may be dithered with a peak dither level of around -130 dB
Exporting to 16-bit PCM WAV format may be dithered with a peak dither level of around -80 dB

When exporting to 16-bit, the maximum "error" is +/- 3 on a scale of −32768 to +32767. 16-bit audio can at best be only accurate to the nearest digit on this scale.

When Exporting to 24-bit, the maximum "error" is +/- 3 on a scale of -8388608 to +8388607. At the present time, state of the art audio electronics exhibit greater inaccuracies than are produced by dithering to 24-bit format."
Old 9th January 2017
  #49
Lives for gear
 
lectric's Avatar
 

Quote:
Originally Posted by grrrayson View Post
Also, to everyone: where is a definitive list of which DAWs dither properly (and how do we even know whether they do or not)? It seems we should make a running one that can be updated over time.
Old 9th January 2017
  #50
Motown legend
 
Bob Olhsson's Avatar
 

The key point to understand is that truncation without dither causes distortion that is more audible than the noise from dithering. Dither prevents truncation distortion and loss of low level information. It doesn't merely mask the distortion. Even a lot of software developers don't understand this which is why there is endless misinformation circulated about the subject.
Old 10th January 2017
  #51
Lives for gear
 
billcarroll's Avatar
 

When should you use dither ?
SEPTEMBER 13, 2012 BY IAN SHEPHERD

When should you use dither ? - Production Advice
Old 10th January 2017
  #52
Lives for gear
 
Brian Campbell's Avatar
 

I've read a lot of discussions on dither and I'm still confused. I'm running Pro Tools native 24/48k (so floating point internal). I typically print stems in the same session, drums, bass, strings, etc. I take from these discussions that I should have a dither plug on the master before each of those print tracks and on the master that's feeding the DAC going to the monitors. Is that correct? Some say yes some say no. Which is it? The only dither plugs that come with Pro Tools have settings for 8,16 and 20 bit, no 24 bit. Why no 24 bit?
If this is a necessary part of using a floating point DAW then why the hell isn't the dither built into the damn thing in the first place, or is it?
I've only dithered when going to 16 bit.
Old 10th January 2017
  #53
Motown legend
 
Bob Olhsson's Avatar
 

All good questions!

I've begun mixing into other tracks at 32 bit floating point and then I play the track through a dither plug-in. Kazrog's masterDither seems to be good and it has no latency unlike the various dynamics plug-ins that include 24 bit dither.
Old 10th January 2017
  #54
Like Bob, I do all intermediate processing and stems in floating point. Therefore, the only times I dither are when outputting a final mix to a fixed-point format, and when using a hardware insert on my ITB mix. In the first case I tend to use something like Pow'R 2 or Pow'R 3 if it's a 16-bit destination, and plain old TPDF dither for 24-bit outputs. Hardware inserts always get TPDF because one mustn't do further processing on noise-shaped dither.

Of course the feed to my 24-bit control-room DAC is always dithered. I try to use the same dither I'll be using at the end. Sequoia allows me to truncate my CR DAC feed to 16 bits by checking a box, and I can enable Pow'R dither on this output if desired. Thus, I can mix using the same settings that I'll use for final output.

Are there really still DAW's that process in floating point or 48-bit+ fixed point, but save intermediate results in 24 bit files? Hopefully, they are doing proper dithering by default before saving, but it's best to ask. Make sure there's some kind of dither, even if you have to do it yourself!

On noise built-up: TPDF adds an extra 4.8 dB of noise (vs. truncation noise) the first time you do it, but the second time it adds only 3 dB; to add an addional 3 dB you have to dither four times total, and eight times for the next 3 dB increment. So the first time you add dither is the worst penalty, but it's a small price to pay: the perceptual equivalent of undithered truncation is about 12 dB worse because distortion is so much more audible than noise.

Still, if you're worried about that first 4.8 dB penalty, and your destination word length is 24 bits, you could cut it down to 3 dB by doing your bounces with RPDF dither instead. RPDF dither eliminates truncation distortion just as well as TPDF dither, but it doesn't prevent the signal from modulating the noise floor. However if your signal level is close to full scale, then your (digital) noise floor is roughly 140 dB down, so who cares if it gets modulated? Ya ain't gonna hear it! OTOH, ya ain't gonna hear it at 138 dB down either, so just turn on TPDF and stop worrying about it!

David L. Rick
Seventh String Recording
Old 10th January 2017
  #55
Motown legend
 
Bob Olhsson's Avatar
 

Most important, the dither is random while the truncation distortion adds up something like 6dB. a pass.
Old 10th January 2017
  #56
Gear Addict
Quote:
Originally Posted by Bob Olhsson View Post
Dither prevents truncation distortion and loss of low level information. It doesn't merely mask the distortion.
This is a great point. Since we expect quantization errors as a byproduct of conversion, dither software can be programmed to tell the error at what frequency and amplitude to occur (to an extent). It can be programmed to become inaudible by shifting it to known simultaneous/temporal masking levels and/or by shifting it to a less sonically-sensitive area of the frequency spectrum.
Old 10th January 2017
  #57
Gear Maniac
 

Quote:
Originally Posted by inestima View Post
This is a great point. Since we expect quantization errors as a byproduct of conversion, dither software can be programmed to tell the error at what frequency and amplitude to occur (to an extent). It can be programmed to become inaudible by shifting it to known simultaneous/temporal masking levels and/or by shifting it to a less sonically-sensitive area of the frequency spectrum.
Yeah, there should be like "mp3 psychoacoustic masking technique" dither.
Old 10th January 2017
  #58
Lives for gear
 
JulenJVM's Avatar
I have asked this in another thread, but the doubt is eating my guts. What about when you´re compiling tracks (editing different takes into 1 final take), with no other processing other than fades and crossfades? Is dither necessary in those cases, when producing 24 bit files on a 32 bit float DAW? Apparently, the audio stays at 24 bit if there is no processing or no gain changes, so there´d be no need to dither. But again, a fade is somehow a gain change (<-- this is what does my head in)
Old 10th January 2017
  #59
Lives for gear
 
Brian Campbell's Avatar
 

I think it's finally 'sinking in'.
Running floating point keep a dither plug on the monitor buss.
When I'm delivering multiple 5.1 stems, 5.1 master, LoRo, (24/48k) for dubbing, probably a safe bet to dither to all those print tracks rather than messing around printing 32 bit float.
'Plain old TPDF' for 24 bit and it's ok to do it more than once.
Some flavour of noise shaping when going to 16 bit (I've always done this).

The Kazrog MasterDither looks alright, 20 bux.
Old 10th January 2017
  #60
Motown legend
 
Bob Olhsson's Avatar
 

I don't know about predictive quantization errors however "smart" dither that turns off when not needed is pretty common. Too bad every DAW doesn't simply do that. We should really never need to think about dither.

Masking varies greatly from individual to individual so it's never a good idea.
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump