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 9th March 2019
  #841
Lives for gear
Quote:
Originally Posted by goat76 View Post
I really liked what the NJAD did for the bass but got the impression that I lost some depth in the mix.
I feel very skeptic about such statements. Did you attempt to do a blind ABX test? If your impressions checks out I feel wary about these dithering algorithms.

Did you also try one of the more common ones?
Old 9th March 2019
  #842
Definitely just a case of "go for it".

Technically, yes, you're adding a second layer of unnecessary noise... at about -144 dBFS - so if you got your mix loud enough it really doesn't matter. And if you, on the other hand, are actually hearing benefits out of the combination, there's no need to think about it anymore.

I'm wondering what the developer of both plug-ins @ chrisj has to say about it though.

Lic. Ezequiel Morfi | TITANIO
Buenos Aires, Argentina.-
Old 9th March 2019
  #843
Here for the gear
 

Quote:
Originally Posted by Mikael B View Post
I feel very skeptic about such statements. Did you attempt to do a blind ABX test? If your impressions checks out I feel wary about these dithering algorithms.

Did you also try one of the more common ones?
Just some info on my project. My songs are stereo files rendered to 64 FP for mastering in Reaper.

I have not done any blind test but it's obvious that the NJAD makes the bass content pop out, the bass guitar takes one step closer in the mix and gets a little more gritty (in a good way that is).
I'm less certain about the depth thing I said, maybe it just gets covered a little by the stronger bass?

If I understand it correctly, the Airwindows NotJustAnotherDither contains noise shaping and the Naturalize Dither is a "just" a more straight forward dither without the N/S. I guess that's the thing I'm hearing when I flip them on/off.

I thougth the Airwindows dithers are common, which ones do you mean?
Old 9th March 2019
  #844
Lives for gear
 

Quote:
Originally Posted by Mikael B View Post
I feel very skeptic about such statements. Did you attempt to do a blind ABX test? If your impressions checks out I feel wary about these dithering algorithms.

Did you also try one of the more common ones?
Fine to be ...

You could test your 'wariness' on your material, as the AIRWindows plugins are freely available. [There is no INSTALL issue, as they are simple DLL files to drop in your plug folder [VST]. If you don't like ... delete it.
Old 9th March 2019
  #845
Lives for gear
Quote:
Originally Posted by goat76 View Post
I thougth the Airwindows dithers are common, which ones do you mean?
Well, like "the industry standard technically correct boring dither": TPDF Dither. or Triangular Probability Density Function, which I'd assume is one of the dither options in your DAW. I haven't really checked that detail though.

Last edited by Mikael B; 10th March 2019 at 12:45 AM..
Old 9th March 2019
  #846
Lives for gear
Quote:
Originally Posted by RJHollins View Post
Fine to be ...

You could test your 'wariness' on your material, as the AIRWindows plugins are freely available. [There is no INSTALL issue, as they are simple DLL files to drop in your plug folder [VST]. If you don't like ... delete it.
That's obvious. It doesn't matter if I like it. I could tell myself many imaginative things and waste my time.

I'd rather attempt to use something like A New ABX Tool | Homebrewed Music.
Old 9th March 2019
  #847
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by goat76 View Post
I have a question for all of you dither experts.

I played around with the Airwindows NotJustAnotherDither and the NaturalizeDither to decide which one I prefer in my project. I really liked what the NJAD did for the bass but got the impression that I lost some depth in the mix.
NaturalizeDither did not alter the depth and didn't change the sound much but when I by mistake had both on at the same time it sounds like I got the best of them both, good bass and the depth was still there, I think.

Maybe some of you can tell me this stacking of two dithers is bad practice from a technical point of view and if it will reduce sound quality, or is it just a case of -if it sounds better, just go for it situation?
Both of those are variants of forcing the output to comply better to a Benford Realness calculation, so technically they're not properly dithers at all, just wordlength reducers. The most recent and generally considered best version of this is the version of NJAD in the plugin StudioTan. That's a plugin with a selection of different wordlength reducers totally outside the scope of this discussion (StudioTan and Dither Me Timbers are essentially using quantization to EQ, not to dither) and it's got the most current NJAD: I have not been removing or altering plugins like Naturalize even though they aren't optimal, because people use them and develop attachments to 'em so I try to keep historical stuff available.

These don't count as 'correct dithers', they're something else. The only reason to call them dithers is that they must function in the same context as one: wordlength reduction. Therefore, it's impossibly confusing to call them anything else.
Old 10th March 2019
  #848
Quote:
Originally Posted by chrisj View Post
These don't count as 'correct dithers', they're something else. The only reason to call them dithers is that they must function in the same context as one: wordlength reduction. Therefore, it's impossibly confusing to call them anything else.
Dithers or not, are they still safe to use for a commercial release???

I'm on NJACD (not StudioTan, the stand-alone plug-in) for every master I do...

Lic. Ezequiel Morfi | TITANIO
Buenos Aires, Argentina.-
Old 10th March 2019
  #849
Here for the gear
 

Quote:
Originally Posted by chrisj View Post
Both of those are variants of forcing the output to comply better to a Benford Realness calculation, so technically they're not properly dithering at all, just wordlength reducers. The most recent and generally considered best version of this is the version of NJAD in the plugin StudioTan. That's a plugin with a selection of different wordlength reducers totally outside the scope of this discussion (StudioTan and Dither Me Timbers are essentially using quantization to EQ, not to dither) and it's got the most current NJAD: I have not been removing or altering plugins like Naturalize even though they aren't optimal, because people use them and develop attachments to 'em so I try to keep historical stuff available.

These don't count as 'correct dithers', they're something else. The only reason to call them dithers is that they must function in the same context as one: wordlength reduction. Therefore, it's impossibly confusing to call them anything else.
Thank you for the information, chris!

As I said, I like them both and I used the NJAD alone for the render but I think I must try the StudioTan version as well.
The dither or not to dither question was a bigger topic than I first thought.

Outside your box of plugins, if we talk ordinary dithers, is the Reapers built-in variant good or should I go for something else?
Old 10th March 2019
  #850
Quote:
Originally Posted by goat76 View Post
Outside your box of plugins, if we talk ordinary dithers, is the Reapers built-in variant good or should I go for something else?
It's very, very good. And gives a handful of options as well.

Lic. Ezequiel Morfi | TITANIO
Buenos Aires, Argentina.-
Old 10th March 2019
  #851
Here for the gear
 

Quote:
Originally Posted by morfi View Post
It's very, very good. And gives a handful of options as well.

Lic. Ezequiel Morfi | TITANIO
Buenos Aires, Argentina.-
I guess it's the JS Bit Reduction/Dither you are talking about?
The Reaper built-in dither is just a check-box in the render options.

I must say, I still want to know what's essential for a correct dither.
What is missing from the Airwindows bit reduction plugins to make them proper dithers, and do we really need the missing parts?
Old 10th March 2019
  #852
Gear Guru
 
Karloff70's Avatar
 

Quote:
Originally Posted by mutetourettes View Post
Just to say thankyou to Paul, Chris, David, Fabien, Alexey for making this a real top-quality educational thread!

Get well soon, Paul!
I'd like to echo that!!

Also, without wanting to interrupt the brains here:

Chris, I downloaded new versions of your plugs a couple of days ago, including Ditherbox and my ears are liking the 24bit Naturalize in there on Logic's output more than my previous (before your floating meddle) Pauldither in a noticeable way. Am I to take out of this discussion (only briefly scanned the last few pages, bit over my head) that this means I like added harmonics unrelatwd to the signal, or have I misunderstood where things are at here right now?
Old 10th March 2019
  #853
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by goat76 View Post
I guess it's the JS Bit Reduction/Dither you are talking about?
The Reaper built-in dither is just a check-box in the render options.

I must say, I still want to know what's essential for a correct dither.
What is missing from the Airwindows bit reduction plugins to make them proper dithers, and do we really need the missing parts?
It's not fair to ask me these things: the only reasonable answer I can give is 'I made these to sound better than regular old TPDF dither and this is how I did it'. There are some folks who will insist on only TPDF dither (even there, I can do unexpected things: Bob liked my TPDF but was disconcerted by a half-bit of correlation on otherwise uncorrelated dither, and others have jumped all over its quirks. It's because it adds and then subtracts a random source, producing half a LSB of averaged-out offset and making stuff at the threshold of truncation-into-silence occupy three possible LSB values instead of only two, while still accurately TPDFing: that's definitely what's going on there)

Alexey, who's guided the floating point dither discussion and has his own code for doing it, is I think responsible for mBit+ in iZotope's stuff? Derived from MegaBitMax? Alexey would tell you the same things: the special custom wordlength reducer is for sounding better, and he'd have his own angle on how that's being done. I can't speak to mBit+ but I remember the behavior of MegaBitMax and it's exactly what Bob Ohlsson talks about when he's concerned about noise shapers: MegaBitMax had extremely pristine behavior in key areas of the audio band (my Ten Nines, back in the day, that became my Vinyl 'dither', was able to top it but only in the lowest bass, where MegaBitMax did everything to produce a low noise floor at around 1-2K, much like Pow-R 3).

And every single one of those spit out huge amounts of energy way up top around the Nyquist frequency, where you 'can't hear it': Dave Collins thought my Ten Nines was the worst and strangest dither he had ever heard in his life, because he CAN hear up there. And all that energy will also cause issues with further processing: a concern when your output is seeing digital volume controls, normalizers, and music streaming services with built-in limiters and such. We don't do this high energy noise shaping mid-production, only at the end… though you could easily do something like throw MegaBitMax on stems going to an OTB mix that used gear with nice beefy transformers, to step on the highs in a natural way: you'd get all the midrange benefits and would be controlling highs with basically an inductor. On the other hand, cold transistor gear might become a nightmare in that scenario, with just the sort of intermodulation distortion people complain about when griping about high sample rates being too high.

What I do lately isn't like that. It's doing different mechanisms for quantization itself, and then doing some very basic noise shaping just to manage that. Not as high energy or as fierce about clearing up 'key perception areas', so it's more flexible… 'safe', as it were. But there's still the argument that you should only use TPDF: anything else is picking colors that you like, not being 'correct'.

I let people still pick and use Naturalize even though the current version of THAT tech is the version of NJAD in the plugin StudioTan. That's because people will pick colors they like, and it's really not for me to tell them they're wrong: I made NJAD out of Naturalize by fixing some behaviors that I thought were undesirable, and I made the newer NJAD by tuning it until the background was more black and quiet and without a trebly liveliness the original NJAD had. People responded to the StudioTan version of NJAD positively, and it's the one _I_ like best and I think it sounds better than TPDF dither and doesn't do things that would mess with further processing, but I'm SUPPOSED to think that, because I made it!

You can specify what's maximally accurate for scientific measurements. Alexey Lukin even helped me do that with dithering to floating point, which isn't a concept that existed before we hammered that one out.

Our perceptions aren't scientific measurements, and our environments aren't controlled. We have sensitivities in our hearing and gravitate to different things: the one thing Naturalize and all version of NJAD have in common is using a Benford Realness calculation, effectively distorting the sample stream's low bits to resemble a live audio capture even if that's not what's supposed to be represented by those samples. Forcing a high Benford Realness measurement is distorting and cheating, but I like how it sounds and so do others: there are kinds of regularity that come off as unpleasant artifacts and the Naturalize/NJAD stuff refuses to produce those outputs even if they would be correct representations of the audio stream.

Use your ears and use your own judgement about how far you want to go away from perfectly accurate and correct. We can do perfectly accurate and correct digital math in DAWs, have done for years, and a lot of us think it sounds terrible. If you do 100% digital summing and stock biquad EQs (or phase linear) and no distortions, ever, and dither your 64-bit buss down to your 16 bit output for CD, you're doing things that are widely considered to be if not mathematically perfect, then mighty close (assuming you're really rigorously doing it right)… and then if your output is soulless flat crud and neither you nor anyone else can fall in love with it, you may well start asking which rules you can break to get more mojo happening.

Bring on the Shun Mook magic wood blocks to stick on things, bring on the Dangerous 2-Buss, bring on fancy wordlength reducers that wildly improve performance at 1K or that use Benford Realness calculations or treat quantization as an EQ… and which of these things are real? All of them… if you go to the people that make them and ask 'is your thing real and should I buy one?' They're compelled to side with their thing, otherwise there's something wrong with the creator. If you get guys like me whose brand it is to be ruthlessly forthright about stuff, they will tell you all sides of the issue and where their thing falls down, and STILL they will tell you to use their thing… or they'll design a new one they're happier with.

And sell you it.

Yes you should use all my best dithers and join my Patreon because I am, indeed, trying my best.

Or, do your own experimenting and form your own judgements based on what works, what you care about, whose opinion you respect, and how risk-averse you are. I am the first to admit that I really blow it sometimes, because I am far too experimental. I don't think my dither stuff 'blows it', and there's a good selection (NJAD, Vinyl, Spatialize) that are both original and effective in their own ways, and don't get nasty with post-processing. But you've got to work stuff out for yourself. You can't simply do what you're told when there are so many contradictory people telling you what to do. And I can't break faith with my own stuff when it does what I wanted it to do. I'm a subjective-experience guy who's prepared to iterate on ideas. I've been taking them in a direction, and that doesn't mean everyone else has to come along.
Old 10th March 2019
  #854
Motown legend
 
Bob Olhsson's Avatar
 

Our problem today is that we are no longer at the end of the digital signal processing chain like we were twenty years ago. We need to consider what it will sound like converted to an MP3 and then fed into a digital volume control. In addition, converters have improved a great deal. As they have improved, noise shape dither seems like less of an improvement to my ear that it used to be over mathematically correct TPDF.
Old 11th March 2019
  #855
Lives for gear
 
Arksun's Avatar
Quote:
Originally Posted by goat76 View Post

I have not done any blind test but it's obvious that the NJAD makes the bass content pop out, the bass guitar takes one step closer in the mix and gets a little more gritty (in a good way that is).
Expectation bias is an incredibly powerful thing, I've fallen for it myself plenty times in the past. I would highly recommend blind testing these things, the results may well surprise you
Old 11th March 2019
  #856
Here for the gear
 

Quote:
Originally Posted by Arksun View Post
Expectation bias is an incredibly powerful thing, I've fallen for it myself plenty times in the past. I would highly recommend blind testing these things, the results may well surprise you
Yes, I know. I think all of us has found ourself make subtle changes and hearing it, and after that realize the plugin is off.

But in this case, the change was too obvious when I flip it on/off in an A/B test to just be expectation bias. I did a blind test sort of. I shut my eyes and turned the plugin on/off until I didn't know if I was using it or not and I could still hear it.
I guess it depends on the recorded content on how much you hear it or if you hear it.

If you read the post by Chris, you see that he made his plugins to not sound like ordinary dithers, he made them sound better or more to his liking so I guess that's what I'm hearing.


By the way, thank you Chris for the long reply and for making good plugins.
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