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
Dithering to 24 bit question (64 bit plugins/DAW) Dynamics Plugins
Old 4th June 2016
  #31
Quote:
Originally Posted by monomer View Post
Well, the quote states that it is the mathematically correct to dither floats and fixed point numbers when reducing. So your statement that you cannot dither floats is nonsense according to the text you quote.
You misunderstood the quote. That's not what it is saying (or what the paper is saying).
Old 4th June 2016
  #32
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by reakktor View Post
Only when the scaling becomes fixed (floating point to integer conversion).
But this is the case when rendering to 24 bits, right?
Old 4th June 2016
  #33
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by FabienTDR View Post
You misunderstood the quote. That's not what it is saying (or what the paper is saying)
I don't seem to have access to download the paper so i was going by what the quote said.

Edit: Also, i don't understand what there is to understand about that quote. It's quite explicit.
Old 4th June 2016
  #34
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by monomer View Post
Yeah, that's not really true. As soon as you do any processing the bits of the original file will get mapped to bits in the internal format. Even a small gain change immediately fills up the newy available bits in the internal format. If you then want a 24 bit file at the output again you will need to dither.
Maybe I didnt understood the OP, but I always trust my ears most than wikipedia. I never care about the theory when processing audio when practise can cover that. What I undersrood from thewas op was that if I work with 24 bits files in reaper for example (64 bits processin) at 24 bits project bitrate, I should dither to get a better 24 bits output , and that sounds wrong. If i got it wrong then I said nothing. But I still say that if it sounds good its good, always.
Old 4th June 2016
  #35
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by reakktor View Post
But, you should research that first. They are talking about the number of amplitude points irrespective of peak values...
I researched what I needed to get a good sound. Technical theory I dont care too much. But I answered according to what I thaught the op was refering, and still think theres no need to dither a 24 bit mixdown made of processed 24 file/s, never, if thats what he was talking about.
Old 4th June 2016
  #36
Gear Addict
 

Quote:
Originally Posted by FabienTDR View Post
It's really important to understand that digital audio uses at least two very different forms of representation values. And there are different contexts each having different meanings.

There's the context of an audio app's internal precision. There's the plugins internal precision, and there's also file storage and transfer. All have different implications.


I warmly recommend anyone to grasp this stuff down to the last atom, it's your profession after all. Misunderstanding the canvas is where most trouble begins.

For this specific thread, I wonder why nobody mentioned that we're working with floating point number representation in 99% of the cases. Be it synthesis, processing, mixing, often also storage and transfer (essentially everything except "export for production/distribution").

Now here's an annoying detail:

Floating point signals cannot be dithered (only fixed point signals can).


The quantization distortion produced when floating point signals reach their precision limit is somewhat less problematic than the sound fixed point truncation. Do not try to dither your processing, as it mostly probably bases on floating point representation.

If the source already contains high amounts of noise (i.e. tape, mic, etc), there's really no need to add additional dither before truncation. But it's worth trying out.

Dithering a floating point signal is technically nothing else than an increase of the noise floor. In addition to the quantisation distortion!
so everything was wrong?? If I not mistaken I remember reading a Bob Katz books and Paul Frindle posts in which they mentioned dithering was important even from 32 to 24 bit, but maybe it was in the processing context rather than to storage.
So if I understand what you are saying correctly, at least theorically it would be irrelevant to dither from a 64 or 32 bit audio engine to 24 if it is done at once?

Thanks!
Old 4th June 2016
  #37
No not everything is wrong. But floating point vs fixed point are very different worlds. Dithering is for fixed point only. In case of floating point, the theory isn't correct, and the noise doesn't dither anything.

Put differently, adding noise to fight floating point truncation is not helpful, it just raises the noise floor. Quantization distortion is still there.
Old 4th June 2016
  #38
Gear Maniac
 

Quote:
Originally Posted by Dina Mastering View Post
I researched what I needed to get a good sound. Technical theory I dont care too much. But I answered according to what I thaught the op was refering, and still think theres no need to dither a 24 bit mixdown made of processed 24 file/s, never, if thats what he was talking about.


IMHO, if someone wants to be a mastering engineer (or make a living that way for that matter) then that person should grasp these basic digital principals and very well... or at least know well enough so that it can be explained to their customers if questions arise...

FWIW.
Old 4th June 2016
  #39
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by Dina Mastering View Post
What I undersrood from thewas op was that if I work with 24 bits files in reaper for example (64 bits processin) at 24 bits project bitrate, I should dither to get a better 24 bits output , and that sounds wrong. If i got it wrong then I said nothing. But I still say that if it sounds good its good, always.
Well, the end result may certainly be good enough in either case.

A lot would depend on what type of numbers reaper uses to represent your 24 bit files.

If you mean the 'track mixing bit depth' in reaper then you're right as the actual bitdepth of the system (so all results of any processing) would be 24 bit already. In this case you assume reaper applies any nessesary dither at the right places inside the engine.

But the reaper default is 64 bit floats so information from your original file will move around and migrate to the available bits after any processing. If you then want a 24 bit file from that you should make sure the output gets dithered (i don't know the exact options for this in reaper).
Old 4th June 2016
  #40
Motown legend
 
Bob Olhsson's Avatar
 

Verified Member
Most monitoring D to As are 24 bit fixed so you almost always must switch from float to integer in order to hear anything. The other factor is that truncation distortion is louder and adds up faster than the noise from dither. My experience has been that if the source material has ever been truncated, it's very hard to hear 24 bit or even 16 bit dither but when you start with previously unprocessed recordings from microphones, the lack of 24 bit dither can be perceived as a loss of depth and imaging.
Old 4th June 2016
  #41
Gear Addict
 

Quote:
Originally Posted by FabienTDR View Post
No not everything is wrong. But floating point vs fixed point are very different worlds. Dithering is for fixed point only. In case of floating point, the theory isn't correct, and the noise doesn't dither anything.

Put differently, adding noise to fight floating point truncation is not helpful, it just raises the noise floor. Quantization distortion is still there.
Thanks for the response, so if I were to render a mix done in a 64bit floating point daw down to 24 bit I shouldn't dither, right?
Neither if i were to export it directly from 64 to 16bit? Just truncate the mix down to 16bit?

I don't understand in which case one would use dither then, if a mix is done itb, audio is processed at 64bit floating-point and next step should be 16 bit fixed-point. There should be no dither? Or I am getting it wrong?

(only case I imagine now for dither is if I captured from an analog chain at 24bit and then wanted to go to 16bit)

Sorry for all the questions, thanks for sharing your knowleadge!
Old 5th June 2016
  #42
Quote:
Originally Posted by tk_nacho View Post
Thanks for the response, so if I were to render a mix done in a 64bit floating point daw down to 24 bit I shouldn't dither, right?
Neither if i were to export it directly from 64 to 16bit? Just truncate the mix down to 16bit?

I don't understand in which case one would use dither then, if a mix is done itb, audio is processed at 64bit floating-point and next step should be 16 bit fixed-point. There should be no dither? Or I am getting it wrong?

(only case I imagine now for dither is if I captured from an analog chain at 24bit and then wanted to go to 16bit)

Sorry for all the questions, thanks for sharing your knowleadge!


There are several things here:

A. Floating point to fixed point conversion

This creates minor losses. Keep in mind that the default 32bit floating point format you often see in DAWs have a maximum precision of 25bit in the best case. Not 32bit.

IMHO the best idea before doing this is (maybe surprisingly) normalization.

1. Export these 64bit fp or 32bit fp files for the archive and later edits. You can add 24bit dither if you want, but technically, it doesn't matter much at this scale.

2. Now create a 24bit fixed point file, it allows you to properly dither it for 16bit and truncate to 16bit.

2b. you can also directly add 16bit dither in your project and export to 16bit fixed if you want. But you can't normalize before truncation in this case (which doesn't always make sense depending on your workflows).


More clearly, dither only really helps when truncating fixed point files, or preparing a future fixed point truncation. It makes no sense to dither a 64bit fp to 32bit fp conversion. Or to dither a gain change happening on a floating point signal.
Old 5th June 2016
  #43
Motown legend
 
Bob Olhsson's Avatar
 

Verified Member
The proper dither level for fixed 24 bit files is much lower than that for 16 bits.

By the way, if we didn't have incompetently programmed DAWs, we could simply select an output format and never worry about it. This was all worked out for digital telephones in the 1950s!
Old 5th June 2016
  #44
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
I havent researched it, but Id say the 64 bit is not the audio directly, its the processing resolution if that makes any sense.
It depends on the DAW. Cakewalk Sonar for instance uses a 64 bit float signal path throughout and sums at 64 bit float. Pro Tools (since PT 11 I believe) uses a 32 bit float signal path but sums at 64 bit float. Cubase uses a 32 bit float signal path and also sums at 32 bit. So it varies a bit.

For processing it depends on the DAW, the processing in question and/or the 3rd party plugins used.

Most DAWs have user settings for the bit depth of recorded, imported or rendered files.

Alistair
Old 5th June 2016
  #45
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by reakktor View Post
IMHO, if someone wants to be a mastering engineer (or make a living that way for that matter) then that person should grasp these basic digital principals and very well... or at least know well enough so that it can be explained to their customers if questions arise...

FWIW.
Thanks for the free unsolicited advice. I didn't expect so much care for what's good for me from a total stranger. But I know and researched exactly what I needed to make my masters sound good and to make my clients happy. Not once since I started I felt the need to dither a 24 bits file that was originally at 24 bits, or had the need for this specific techical theorical knowledge to make my masters sound good.

And this subject, if we are talking about the same, its' not the basics of audio for me at all, it's technical specifycs about how daws/plugins work internally. I'm not a plugin / daw designer. You don't need to know how a pickup works to play an electric guitar. I hardly doubt I'll never need to tell my clients about how their daw works internally. If they ask me if they have to dither a 24 file, wich I doubt they'll do, for now I'll tell them to not do it.

Anyway, the next time I optuput a 24 file from a 24 file I might try to dither it. If it sounds better I'll dither, if not I'll don't. I don't need to know why it will sound better or worse to have it sound better or worse, or anybody to tell me if he thinks I am or I am not a mastering engineer for not caring about something that will most likely not affect the the way my work sounds or I don't need to understand specifically to benefit from it.

Until I try, I still think it makes no sense to dither a 24 file to a 24 output on a 24 project, and I don't think too much people do, but who knows, maybe I'll discover audio nirvana when I start doing it.
Old 5th June 2016
  #46
Lives for gear
 

Verified Member
Quote:
Originally Posted by monomer View Post
But this is the case when rendering to 24 bits, right?
Yes,

Fabien isn't making it clear that dither is about the DESTINATION format, not the source.

If your destination is fixed point, and your source has information which is higer resolution than the destination (as will be the case in a 32 bit float to 24 bit fixed conversion, and especially so in the case of 64 bit float to 24 bit fixed), then strictly speaking, the correct way to do that conversion is with dither.

Does it make any audible difference? I find it hard to envisage real world circumstances where quantization distortion at -140dBFS could become audible, but I'll remain open to the possibility.

But, here's an important thing. Assuming it's done right (and it's really not difficult to do right), IT CERTAINLY WON'T DO ANY HARM.

So I say stop pontificating about it, and just do it.
Old 5th June 2016
  #47
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
Until I try, I still think it makes no sense to dither a 24 file to a 24 output on a 24 project, and I don't think too much people do, but who knows, maybe I'll discover audio nirvana when I start doing it.
Unless you don't do anything at all to the file, you are not going from 24 bit to 24 bit though. You are starting at 24 bit, then going to the internal resolution of your DAW (and/or whatever processing you are applying including gain changes) and then back to 24 bit. That is why technically dithering should be applied.

If you are not doing anything at all to the files then there is no need to dither, no. Keep in mind that even a simple gain change is a change and you are no longer going from 24 bit to 24 bit. There is an intermediary step that is at a different bit depth.

Alistair
Old 5th June 2016
  #48
Gear Maniac
 
Dina Mastering's Avatar
 

Ok, theory aside I just did a quick test on 24 mix I'm working on.

I compared a 24 to 24 render with dither VS without dither (in a 24 project with extensive processing) and the difference was negligible, and I couldn't decide on one in repeated blind AB, at least with this mix. Maybe with a several mixes test, but I think for the difference I guess it would make, if any, I'll pass for now. Maybe someday.

I also compared a direct render to 16 bits dither VS a dithered reduction to 16 of the two previous renders, the 24 dithered and the 24 not dithered. In this case I felt it was clearer that the worse version was the one with a dither to 24 and a dither to 16, and the best probably the direct to 16 or tie in between this and the 16 dithered from the 24 undithered (should check again).

So I guess for me the possible micro advantage of dithering a 24 bit to 24 bit output is clearly supprassed by the quite possible slightly noticeable loss of a further reduction to 16 bits. So yeah, no dither to 24 for me for now. And yes, a minimal tweak of an eq band would make the file quite more different than the differences we are talking here. YMMV
Old 5th June 2016
  #49
Motown legend
 
Bob Olhsson's Avatar
 

Verified Member
Look for better 24 bit dither!

Ultimately the issue is how crunchy additional processing such as consumer digital volume and tone controls make the music.
Old 6th June 2016
  #50
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
Look for better 24 bit dither!
So a different noise shape will change the results?

Ok, wich one you recomend?
Old 6th June 2016
  #51
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
So a different noise shape will change the results?

Ok, wich one you recomend?
You could try flat dither or very shallow noise shaping when going to 24 bit.

The point of noise shaping is to push the noise away from the frequencies we are most sensitive to but when it is at around -140 dB FS this is less important.

I don't know which dither you normally use but I seem to get good results with iZotope's MBIT dither with the shaping set to moderate settings.

Alistair
Old 6th June 2016
  #52
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by UnderTow View Post
I don't know which dither you normally use but I seem to get good results with iZotope's MBIT dither with the shaping set to moderate settings.
That's exactly the one I used for my test.


Have you people actually made any tests and heard the difference you are talking about for yourselves? Or you're just talking about the theory? Because I'd say most people all over the internet and here at gearslutz seems to agree that the difference in between most good dithers is quite negligible in most practical situations, and I think a lot of people just dither with their choosen dither only when they reduce bith depth.

Again, if I choose to use another good dither, I think the difference would be less noticeable than boosting 0.1 db more or less in an eq band.

We all want the best sound possible, but I'm under the impression that most people prefer to spend more time with eq or comp than choosing a different dither.

And after my test, I still think I'll keep not dithering my 24 bit files unless I'm totaly sure the audio will never be changed or bit reduced at some point for the reasons I said before, or maybe not even then. My test works for me and my setup YMMV

Last edited by Dina Mastering; 6th June 2016 at 12:27 PM.. Reason: Removed parts because I thaught I was answering to another person.
Old 6th June 2016
  #53
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
That's exactly the one I used for my test.


Have you people actually made any tests and heard the difference you are talking about for yourselves? Or you're just talking about the theory? Because I'd say most people all over the internet and here at gearslutz seems to agree that the difference in between most good dithers is quite negligible in most practical situations, and I think a lot of people just dither with their choosen dither only when they reduce bith depth.

Again, if I choose to use another good dither, I think the difference would be less noticeable than boosting 0.1 db more or less in an eq band.

We all want the best sound possible, but I'm under the impression that most people prefer to spend more time with eq or comp than choosing a different dither.

And after my test, I still think I'll keep not dithering my 24 bit files unless I'm totaly sure the audio will never be changed or bit reduced at some point for the reasons I said before, or maybe not even then. My test works for me and my setup YMMV
I don't hear dither at 24 bit if I have my monitoring at anything but insane levels. But my thinking is exactly the opposite of yours: Because I don't know what might happen down the line I stick to best practices and share the philosophy that Jon posted about above.

What I have done besides listening tests is do extensive technical tests. If you don't dither to 24 bits you will have quantization distortion that is _higher_ than the level of proper dithering. You might not hear it but skipping dithering is measurably worse than proper dithering. So why not simply dither?

Alistair
Old 6th June 2016
  #54
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by UnderTow View Post
I don't hear dither at 24 bit if I have my monitoring at anything but insane levels. But my thinking is exactly the opposite of yours: Because I don't know what might happen down the line I stick to best practices and share the philosophy that Jon posted about above.

What I have done besides listening tests is do extensive technical tests. If you don't dither to 24 bits you will have quantization distortion that is _higher_ than the level of proper dithering. You might not hear it but skipping dithering is measurably worse than proper dithering. So why not simply dither?

Alistair
That's nice, to each his own.

In my case, I'll stick to what I heard and probably choose to avoid the possibility of a "redither" that, for me and with my setup/test, it did more clearly audible damage than the "inaudible"/negligible improvement of dithering the 24 file.
Old 6th June 2016
  #55
Lives for gear
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
That's nice, to each his own.

In my case, I'll stick to what I heard and probably choose to avoid the possibility of a "redither" that, for me and with my setup/test, it did more clearly audible damage than the "inaudible"/negligible improvement of dithering the 24 file.
There is no such thing as "redithering".

Dither is just part of a correctly implemented fixed point quantizer, it eliminates distortion AND allows detail smaller than a quantization step to be preserved. It takes away nasties, it doesn't add tham. if switching dither on actually causes any damage, something is broken.
Old 6th June 2016
  #56
Gear Guru
 
UnderTow's Avatar
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
That's nice, to each his own.
Actually, no. Not to each their own. This is maths. There is a correct way to do things and there is an incorrect way.

Quote:
In my case, I'll stick to what I heard and probably choose to avoid the possibility of a "redither" that, for me and with my setup/test, it did more clearly audible damage than the "inaudible"/negligible improvement of dithering the 24 file.
See Jon's response.

Alistair
Old 6th June 2016
  #57
Lives for gear
 
Tarekith's Avatar
 

Verified Member
Quote:
Originally Posted by Bob Olhsson View Post

By the way, if we didn't have incompetently programmed DAWs, we could simply select an output format and never worry about it. This was all worked out for digital telephones in the 1950s!
That's what I've always thought. It seems silly that we have to worry about something like this when it should just be handled automatically.
Old 6th June 2016
  #58
Motown legend
 
Bob Olhsson's Avatar
 

Verified Member
Lack of dithering is actually far more audible to people having hearing damage!
Old 6th June 2016
  #59
Gear Maniac
 
Dina Mastering's Avatar
 

Quote:
Originally Posted by Jon Hodgson View Post
There is no such thing as "redithering".

Dither is just part of a correctly implemented fixed point quantizer, it eliminates distortion AND allows detail smaller than a quantization step to be preserved. It takes away nasties, it doesn't add tham. if switching dither on actually causes any damage, something is broken.
I meant "adding noise twice", but even if that is not technically accurate, what I care is what I tested, and my conclusion from that test was that adding the dither made almost no difference to the 24 file, but a further dithered reduction to 16 bits sounded worse to me using the 24 bits dithered file than the undithered one. I'm not saying people should or shouldn't do, I'm just saying what I'll do.

Quote:
Originally Posted by UnderTow View Post
Actually, no. Not to each their own.
All right, as you say then.

Quote:
Originally Posted by UnderTow View Post
This is maths.
I don't work with numbers, I work with sound.

Quote:
Originally Posted by UnderTow View Post
There is a correct way to do things and there is an incorrect way.
That's exaclty the oposite idea to the one that allowed humanity to evolve.


/ / / / /

Some people here are saying you should dither a 24 bit mixdown from a 24 bits project with 24 bits files when working with a daw that uses different internal processing resolution because the theory says that's the optimal way to do it. That's honestly new and interesting to me. I respect people's reasons to do that and never said I don't or what people should do for sure, but If somebody doesn't respect my decision to do not, specially when I said it's based on my own conclusions from my own practical test with my own setup (so "what works for me") that's certainly not my problem. But hey, best wishes anyway.
Old 6th June 2016
  #60
Lives for gear
 

Verified Member
Quote:
Originally Posted by Dina Mastering View Post
I meant "adding noise twice",
Typically you're not really adding noise to the final product, but rather changing its colour.

Quantization introduces errors, which depending on the signal being quantized can be more or less correlated, the more correlated it is, the more like distortion it sounds (if it's loud enough to hear at all), the less correlated (i.e more random) it is, the more it is just like white noise.

While the implementation of a dithered quantizer involves injecting some noise, this should be viewed in the context of a scenario where noise is being introduced anyway, the net result is that it's more useful to view the complete process as an smoothing out of what you would otherwise get.


Quote:
but even if that is not technically accurate, what I care is what I tested, and my conclusion from that test was that adding the dither made almost no difference to the 24 file,
That surprises me not at all.

Quote:
but a further dithered reduction to 16 bits sounded worse to me using the 24 bits dithered file than the undithered one.
Either something was wrong, or cognitive biases came into play. The worst case scenario for dither is silence, where it would result in about a half bit's worth of broadband noise... in other words white noise at -141dB or so. No way that's going to make a 16 bit quantization (which itself will introduce noise at about -93dB) sound worse.
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
Similar Threads
Thread
Thread Starter / Forum
Replies
plus6vu / Mastering forum
41
JAY55 / Mastering forum
20
hackenslash / Mastering forum
4
sound_forward / So much gear, so little time
4

Forum Jump
Forum Jump