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 14th January 2017
  #91
Lives for gear
 

Quote:
Originally Posted by craigb View Post
Ok, thanks. What bends my brain here is that truncation doesn't produce simple harmonic distortion, and that the distortion produced is a product of a process that is not musically related to the original signal. I guess that just makes it ugly distortion. I'll leave it at that and make sure the dither is on.
Ugly it is! :-)
Old 14th January 2017
  #92
Quote:
Originally Posted by Audiop View Post
Ugly it is! :-)
Well, it certainly is successfully being used for creative purposes!

No idea why, but it seems that most quality hip hop is not dithered. I hear quantization effects all over the place.

https://www.youtube.com/watch?v=0gEPxgPQmRc
https://www.youtube.com/watch?v=s4iR668Ki3I
https://www.youtube.com/watch?v=gVs3oWeRty8
Old 14th January 2017
  #93
Motown legend
 
Bob Olhsson's Avatar
 

The problem is that both dither and truncation distortion accumulate upon repeated calculations. Taking just one truncation event out of context and pronouncing it to be audible or inaudible is meaningless.
Old 14th January 2017
  #94
Lives for gear
 

This thread is useless without pics..

Digitally generated single sine and dual sines at close to full level.

Difference from truncating to 16 bit with and without dither.

I tried to generate a 32 bit file to show truncation to 24 bit but I think the software I have generate a 24 bit file with zeros padded when 32 bit is chosed (basically no diffrence when truncating 32 to 24).

I do think the green trace in pic 1 and 3 is more or less representative of truncating from 32 or 64 bit wordlength to 24 bit.
Attached Thumbnails
Good dither practices, what are yours?-1-truncation24-16.jpg   Good dither practices, what are yours?-2-dither.jpg   Good dither practices, what are yours?-3-trunc_dual.jpg   Good dither practices, what are yours?-4-dual_dither.jpg  
Old 14th January 2017
  #95
Another graph

Quote:
Originally Posted by Audiop View Post
I tried to generate a 32 bit file to show truncation to 24 bit but I think the software I have generate a 24 bit file with zeros padded when 32 bit is chosed (basically no diffrence when truncating 32 to 24).
I posted another graphical example years ago in this thread: Why We Dither

In my example, I think the original signal is floating point, and the output signals are 16 bit fixed point.

David L. Rick
Seventh String Recording
Old 14th January 2017
  #96
Quote:
Originally Posted by Audiop View Post
I was going to say Transfer Function but that is a term is more strictly applicable for a linear time invariant system I've learned. The correct term for a nonlinear system escapes me for the moment.
Nonlinear systems are most generally modeled by means of a Volterra Series expansion. The first term in that expansion is the conventional frequency response. The second term gives the response to pairs of frequencies, the third to combinations of three sinusoids, and so on.

Modest nonlinearities can be modeled more simply by expressing the input-output amplitude curve using a Taylor series. The size of the second order coefficient tells you how big the second order sum and difference frequencies will be, and similarly for the third and higher terms.

David L. Rick
Seventh String Recording
Old 14th January 2017
  #97
IMHO, the essential difference between properly dithered PCM systems and a not properly dithered systems is that the former is a linear system with noise, and the latter is a nonlinear system with noise.

The point that you can literally change this fundamental characteristic by dither is an impressive fact. Both classes of systems behave fundamentally different, with the nonlinear system suffering from notable drawbacks (irreversibility, order of operation matters, no spectral integrity, etc).

Of course, the relevance is rather small with 24bit to 16bit truncation of audio material. At this scale, it's more a matter of principle. But it grows substantially in longer chains, batch processing and SRC in particular (which might well use fixed point between each processing instances). And of course when working with low bit depths.
Old 14th January 2017
  #98
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Audiop View Post
This thread is useless without pics..

Digitally generated single sine and dual sines at close to full level.

Difference from truncating to 16 bit with and without dither.

I tried to generate a 32 bit file to show truncation to 24 bit but I think the software I have generate a 24 bit file with zeros padded when 32 bit is chosed (basically no diffrence when truncating 32 to 24).

I do think the green trace in pic 1 and 3 is more or less representative of truncating from 32 or 64 bit wordlength to 24 bit.
Here my examples:
1. modeled sine 24 bit to 16 bit
2. modeled (without dithering) sine 16 bit to 24 bit with dithering.

What is dithering audio?

For case 1 dithering blur harmonics by quantization.

For case 2 dithering noise mask the harmonics.

Case 2 may be usable for remastering 16 bit stuff to higher bit-depths.

However, it is not effective way for each of 16-bit original.
Old 14th January 2017
  #99
Lives for gear
 
JP__'s Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
The truncation also removes low level detail that is audible through the dither. If the audio has ever been truncated, the differences become much more subtle without that low level detail and the masking from the distortion.
In my experience there seems to be also a relationship between truncation distortion and the distortion of the DAC feeding analog loop. With some the truncation distortion becomes much more obvious than with others.
Edit: just see you mentioned this in your next post.
Old 14th January 2017
  #100
Motown legend
 
Bob Olhsson's Avatar
 

A real problem has been poorly implemented dither. It's critical that dither is made very random between each audio channel. Another problem has been applying dither but leaving actual truncation to 16 bits for when the file is actually written. This leads people making comparisons to mistakenly believe that they are hearing truncation without dither when they actually are comparing 16 bit dither added to or left off of a 24 bit feed.

It's unfortunate that most sources of information depend on advertising for support so audio manufacturers and developers are given a knee-jerk pass instead of having their competence questioned when people hear a problem.
Old 14th January 2017
  #101
Gear Addict
 

Pretty much the simplest, best explanation of dither I've ever seen is in this video (you can skip directly to the dither chapter) - https://www.xiph.org/video/vid2.shtml

The whole video is great actually, it's a super accessible, easy to watch explanation of some first principles of digital audio that always end up being the subject of forum debate...
Old 14th January 2017
  #102
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
A real problem has been poorly implemented dither. It's critical that dither is made very random between each audio channel.
Sorry, I don’t understand what is random between each audio channel. Could you give more details?
Old 14th January 2017
  #103
Motown legend
 
Bob Olhsson's Avatar
 

If it isn't completely random between channels, there is a noticeable loss of imaging.
Old 14th January 2017
  #104
Deleted User
Guest
Reading this is making my head spin. Many well articulated and differing points of view.

Can someone answer these questions for me?

I am using Cubase. Currently, the project setup is configured to create 24-bit files. I can easily reset it to create 32-bit files.

As we know, it works (internally) at 32-bit FP.

Since my 2-buss goes through two virtual outputs (via my UAD Apollo) into Wavelab, which is creating a 32-bit file, should I:
1/ need to dither the output of Cubase
2/ change to 32-bit files, or
3/ not worry about dithering until I go to 16-bit in the master?

There is no analog anywhere in this chain, except for the A/D (going in to Cubase): no hardware inserts or processing. It's completely ITB.

Final question: is there an advantage to letting Cubase just create a 32-bit output file, rather than playing the 2-buss feed into Wavelab in realtime?
Old 14th January 2017
  #105
Quote:
Originally Posted by Yuri Korzunov View Post
Sorry, I don’t understand what is random between each audio channel. Could you give more details?
Yuri, it's important that the dithers in each channel be statistically independent. If they are correlated, your ears will pick it out and try to make a (false) image out of it.

David L. Rick
Seventh String Recording
Old 14th January 2017
  #106
Lives for gear
Quote:
Originally Posted by FabienTDR View Post
quality hip hop
Doesnt exist. So no dither needed.
Old 15th January 2017
  #107
Lives for gear
 

I've read a bazillion dither posts over the years. I don't dither. I just don't. I track at 32bit float or 24bit at 48,96, or 192, I don't make cds, I use lots of external hardware in the daw path, I route stems from d/a to a/d of synchronized daw farms, plug ins routed all over the place via external patch bays in the farm, and I just don't think about dither at all.

I will say that other than quick ref mixes, I don't do my own mixing. I send the tracks out to others. Mastering is usually done by Grundman. So maybe they introduce dither considerations in what they do. I don't hear about it and my stuff sounds fine as far as I'm concerned.

Maybe I'm ranting a little. Dither is just one thing I refuse to think about as part of the creative process. Maybe if I were back in 1995 working in 16bit. I dunno.
Old 15th January 2017
  #108
Lives for gear
 

Quote:
Originally Posted by whiteaxxxe View Post
Doesnt exist. So no dither needed.
Sounds like a Donald Trump tweet.
Old 15th January 2017
  #109
Lives for gear
Quote:
Originally Posted by thenoodle View Post
Sounds like a Donald Trump tweet.
Yeah, well, kinda worked for him... :-((

BTW: I got a warning for mentioning Donald the Trump on this politics-free forum. Just saying.
Old 16th January 2017
  #110
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
If it isn't completely random between channels, there is a noticeable loss of imaging.

Quote:
Originally Posted by David Rick View Post
Yuri, it's important that the dithers in each channel be statistically independent. If they are correlated, your ears will pick it out and try to make a (false) image out of it.
Bob, David,

May be there other reason? Theoretically correlated between channels noise can't cause loss quality. I don't know exactly though.
I released my algorithm as uncorrelated between channels, but I didn't have in mind this.
White noise used for dithering should be uncorrelated. Except each its sample assigned for both channels simultaneously.
Old 16th January 2017
  #111
I only use it for CD prep. Everything else is analog or 24 bits.

"A man's got to know his limitations" ~ Dirty Harry
Old 16th January 2017
  #112
Motown legend
 
Bob Olhsson's Avatar
 

Quote:
Originally Posted by Yuri Korzunov View Post
Bob, David,

May be there other reason? Theoretically correlated between channels noise can't cause loss quality. I don't know exactly though.
I released my algorithm as uncorrelated between channels, but I didn't have in mind this.
White noise used for dithering should be uncorrelated. Except each its sample assigned for both channels simultaneously.
Whose theory?
Old 16th January 2017
  #113
Motown legend
 
Bob Olhsson's Avatar
 

Quote:
Originally Posted by Jim Williams View Post
I only use it for CD prep. Everything else is analog or 24 bits...
It's only 24 bits if there is absolutely zero signal processing or gain changes as if it were an analog tape machine. Otherwise it requires dither to not sound crunchy with a loss of low level detail.
Old 16th January 2017
  #114
Gear Guru
 
Karloff70's Avatar
 

Bob, you make me so happy I started this thread!

*sits back down to listen more*
Old 16th January 2017
  #115
Lives for gear
 
Fidelis's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
It's only 24 bits if there is absolutely zero signal processing or gain changes as if it were an analog tape machine. Otherwise it requires dither to not sound crunchy with a loss of low level detail.
So I did some tests on Pro Tools HD version 12.6.1. I used the Bitter plugin that measures the bit depth of the channel output. So, it works exactly as Bob says. If no plugin processing, volume or pan is done, it measures 24 bit on the output. if I change any thing (0.1dB on the mixer fader) it goes to 32bit (or 64bit).

See pictures.
Attached Thumbnails
Good dither practices, what are yours?-just-volume-change-pro-tools-mixer.png   Good dither practices, what are yours?-no-processing.png  
Old 17th January 2017
  #116
Motown legend
 
Bob Olhsson's Avatar
 

Basically, if any signal processing or gain changes are involved, I bounce to either 96x32 float or 96x24 with dither. Then I sample rate convert to 44.1x32 float or 44.1 24 and do my final processing before reduucing to 44.1x 16. If you aren't processing after sample rate conversion, you could go right down to a 16 bit file but I always leave myself the option to do additional processing or gain changes.
Old 17th January 2017
  #117
Quote:
Originally Posted by chosen one View Post
which dither type is recommended, one or several different types? (waves, dmg, kazrog, airwindows etc all have different option/names for noise shapings and noise types making this more confusing)
For anything that's going to receive further dither or conversion,
I think it's better to avoid noise shaping, and/or if noise shaping is used,
the very last spot (mastering) would work best ime/o. gl
Old 17th January 2017
  #118
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
Whose theory?
I'm not scientist. I'm engeneer.

When I worded under dithering algorithm, I used general dither theory: adding noise for decorrelation rounding error harmonics with musical signal generating it.

If we add similar noise to both channels in signal it will looks like:

1ch: Ch1i=round((Mus1i+Ni)/MaxAmpOrig*2^15)
2ch: Ch2i=round((Mus2i+Ni)/MaxAmpOrig*2^15)

where

Ch1 and Ch2 are resulting dithered signal,

Mus1 and Mus2 are original signal with high bit resolution

N is noise

MaxAmpOrig is maximal amplitude of oroginal signal Mus.

i is number of current sample.

If look to the formula, I can't see what is difference with formula

1ch: Ch1i=round((Mus1i+N1i)/MaxAmpOrig*2^15)
2ch: Ch2i=round((Mus2i+N2i)/MaxAmpOrig*2^15)

where N1 and N2 uncorrelated.

Main my target of development was minimal noise level.

I implemented the algorithm as me was suitable from programming point of view. There should not be correlations between channels. But as old programmer I should add: "if all work as I think"

However I'm not implemented uncorrelated noise delibiratelly.

I understood term "image" as position X, Y, Z of played back sound source for listener.

May be we understand the term differently?
Old 17th January 2017
  #119
Motown legend
 
Bob Olhsson's Avatar
 

Yuri,

I'm talking about not allowing the dither to be somewhat identical between the different channels. This would have to do with exactly how the random numbers called for are actually being generated. Pseudo-random is supposed to be common.

Something else to consider when rounding (I'm remembering all of this from observing conversations 25 years ago.) is that I understand computer FPU routines default to bankers' rounding which is important for financial calculations but should be disabled for accurate signal processing calculations. I believe the problem created is a DC offset.
Old 17th January 2017
  #120
Motown legend
 
Bob Olhsson's Avatar
 

Kazrog does 16 bit and has no latency. You probably want TPDF.
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