Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[OatBran]

I know I know... this is 3 topic posts inside of a week... but I feel these are relatively important discussion points.

Anyhow... I've been playing with and experimenting with various dithering methods and am curious as to what all of you consider to be pertinent variables in determining what to use.

I couldn't help but notice that Mr. Brad Blackwood endorses Crane Songs CD dither!  Interesting.  I'd be interested in hearing it or atleast some material which used it.  With the crowds of simple and complex dithering schemes out there, I was curious which you use, for what reasons, and their implications.

Any takers?

[dcollins]

OatBran wrote on Fri, 21 May 2004 22:55

I couldn't help but notice that Mr. Brad Blackwood endorses Crane Songs CD dither!  Interesting.  I'd be interested in hearing it or atleast some material which used it.  With the crowds of simple and complex dithering schemes out there, I was curious which you use, for what reasons, and their implications.

I suspect that nine times out of ten B-rad doesn't use the Cranesong dither.

As you have probably deduced from reading the "internet," there is no single factor in mastering that will affect the sound more than dither.

Forget spending money on such luxurys as monitoring or acoustics, and start building your collection of different dithers and noise-shapers.  Like the pro's do.

Or, you just forget the whole thing, use a flat or low-order noisemaker and call it a day.

DC

[jfrigo]

My most frequent choice is flat TPDF dither, or perhaps some mild noise shaping if the situation calls for it. The difference between dither flavors is pretty low on the list of things to worry about. Just as long as you use dither, the type is less than an exciting decision. This may be a good time to mention the tech talk page at my website. If you want more in-depth thoughts, have a look. The first couple articles are some years old, but Nyquist & Shannon haven't changed their stripes since I wrote the articles in the 90's. Dither isn't exactly a new topic (though it is to some of course)...

[bblackwood]

dcollins wrote on Sat, 22 May 2004 01:51

I suspect that nine times out of ten B-rad doesn't use the Cranesong dither.

Or even less. On stuff suffering from a bad case of 'digititus' it can be a miracle worker, but for most projects, flat TPDF or POWr work fine...

Quote:

Forget spending money on such luxurys as monitoring or acoustics, and start building your collection of different dithers and noise-shapers.  Like the pro's do.

Excellent point, DC. If your room and monitors are the best you can possibly have and you eeked every bit of performance out of them, then you can start worrying about stuff that even experienced listeners have to hear in fairly controlled environments to discern.

Still never had a client comment on my selections of dithers. Or even ask about it, for that matter...

[jazzius]

It cracks me up every time i read someone over at Gearslutz talking about their favorite flavour of dither.....which they enjoy thru their Mackie monitors......the power of imagintion never ceases to amaze (me).

[TotalSonic]

dcollins wrote on Sat, 22 May 2004 07:51

Forget spending money on such luxurys as monitoring or acoustics, and start building your collection of different dithers and noise-shapers.  Like the pro's do.

Hee hee.  Hey Dave - I think you broke my online sarcasm detector's meters by slammin them to 11 with that one!

It seems what dither does for you is often misunderstood from what I've seen from a lot of general blather on the net.  This is all it will do for you: it smoothes out the sounds of things that fade out to the lowest dynamic levels - so that things like reverb tails instead of fizzling out as they get to the quiestest point blend into the noise floor instead. Good dither should have a few things: randomness, "pleasantness", and should be a tiny bit different on L & R channels (not in level - but in the noise being produced), and it's presence should be not apparent to the listener unless you really focus on it.

So whether the choice of it is critical is kind of dependent on the material and recording. i.e. A typical hip-hop master where the recording was made using 16bit samples & digital reverbs (which tend to fizzle out at the end of their tails anyway), and where the levels are usually pretty constantly slammed (where maybe the only time things fade to zero is at the end of the song) - what you use for dither - heck, whether you even bother to dither at all - really is not going to make any kind of appreciable difference (except for that final fade).  Contrast this with a recording done of say a string quartet done at 24 bit done in a live reverberant room where the material has a serious amount of dynamics and quiet passages - and then dither choice can be much more important as there are all kinds of things whose sound is fading into the distance that you'll want to preserve the natural characteristics of the ambience (which gives a sense of being there in the room) as much as possible.  

What the best thing to use is pretty subjective - and rather than trusting what the "experts" say it's best to judge for yourself.  
There are examples of various dither's noise cranked up so you can hear what the differences are at:  
http://www.24-96.net/dither/
Another way to check this out is to take a 24bit recording of something like, say a single strum on an acoustic guitar that fades to the distance in a really live room.  Process it to 16bit first truncating, and then using all your different dither options that you have. Listen back to the results and see what one you like the best.
Good headphones, good monitors, and a really quiet accurate room is pretty much a necessity for hearing the differences - because they are usually pretty subtle.

But ultimately what eq or comp you choose to set and what box is doing these things will have a much much greater impact on the vast majority of your masters than what dither you choose.

Best regards,
Steve Berson  

[OatBran]

I appreciate the relevant input from those of you who gave it!

I agree that dither is a kind of insignificant consideration. Especially now that converters have come so far from their inception.  In the 80's and early 90's, converters were still pretty bad and dither's need was higher.

However, this being a mastering forum where (I would hope) any details involved with the process can be discussed, I thought I would bring it up.  I found myself with a decision to make in session, however inconsequential, and I wanted to see what others were leaning to.  I was dealing with a 4 piece acoustic set with acoustic guitar, viola, vibes, and an acoustic bass.  In such a situation where dynamics are quite broad my dither choice had become more important.

Once again, thanks for the input.

[bblackwood]

OatBran, all things mastering are open for discussion. For the most part, posters on this board are helpful and stay on topic. I think the point of some of the posts above is that a new guy (not you, perhaps even a lurker), will read stuff about j*tter and dither and not fully understand how minor they are compared to whether or not you add .5dB at 10kHz. Too many people obsess about the little stuff while completely ignoring the major stuff like monitoring and room acoustics. And FYI, I'm quite sure I will always make that point when a discussion of dither and such comes up on my forum...

Again, I'd start with standard TPDF. I never really cared for POWr but needed something beside TPDF not too long ago as a certain project sounded quite a bit flatter at 16 than at 24. I tried POWr and it worked great on that project. Other times I've used POWr and could easily hear the diff on the dithered material, sounded brighter, if not harsher.

[jfrigo]

OatBran wrote on Sat, 22 May 2004 11:59

ecially now that converters have come so far from their inception.  In the 80's and early 90's, converters were still pretty bad and dither's need was higher.

You may have somewhat misunderstood the comments above. Dither is still needed just as much as it was in the 80's, regardless of the improvements in converters. That really has nothing to do with it. The part that was stated as less important is the choice of dither type. It was not stated that the USE of dither is unimportant. Proper dither practice is still quite important.

[OatBran]

Ah!  I see what you were getting at.  However, I do stand by my statement that dither is not as important as it was back in the 80's and early 90's...

Converters then were quite bad as we all know.  Not in the digital end as much as in the analog portions of the circuits.  The digital 'harshness' and the pitiful LSB (least significant bit) detail was absurd.  We were lucky to be getting 12bits of detail and 4 bits of crap.  Most of this was due to our inability at the time to create a steep yet clean anti-aliasing filter among other problems.  Phase issues wrecked the high end and the 96db theoretical dynamic range of 16bit never amounted to more then mid 80-85db at best after all was said and done.

Anywho, back then we had much more to clean up after as far as dither is concerned, IMHO.  This is not to undercut the necessity of dither now by any means.  If I thought dither to be unimportant, I wouldn't have brought this up.

Thanks all.

[Ronny]

What is this dither thing that you guys keep talking about? All that I can find on the internet is that it's a theolodic bombsight design, invented by Jimmy Stewart.

[dcollins]

OatBran wrote on Sat, 22 May 2004 14:01

Quote:

Ah!  I see what you were getting at.  However, I do stand by my statement that dither is not as important as it was back in the 80's and early 90's...

Not true.  When you have poor converter linearity, the dither doesn't work.   The LSB has to be meaningful for the system to do its thing.

Quote:

Converters then were quite bad as we all know.  Not in the digital end as much as in the analog portions of the circuits.  The digital 'harshness' and the pitiful LSB (least significant bit) detail was absurd.

Not really.  It was the converters, and to a lesser degree, the filters.  Take the Sony 1610 for example, Jensen transformers and 990's.  But fed to what?  JVC also had good analog designers.

Quote:

We were lucky to be getting 12bits of detail and 4 bits of crap.  Most of this was due to our inability at the time to create a steep yet clean anti-aliasing filter among other problems.

First, there _were_ A/D converters that came close to 16 bits in the old days, they were just expensive.  Analog Solutions comes to mind....  Second, the LPF's in the form of Murata and later Apolgee, were as linear as seven opamps in series can be...

Quote:

Phase issues wrecked the high end and the 96db theoretical dynamic range of 16bit never amounted to more then mid 80-85db at best after all was said and done.

This is arguable.  I have many, many, CD's that were mastered through brickwall analog filters where the high-end sounds terrific....  I was one of the first people to build custom all-pass filters to correct the HF phase shift of early PCM systems, but we gave up after finding the converter was the problem.

The ear is pretty much deaf to high frequency phase anyway.

Quote:

If I thought dither to be unimportant, I wouldn't have brought this up.

It's not _unimportant_ but just about everything else is more.....

DC

[lucey]

dcollins wrote on Sun, 23 May 2004 00:55

The ear is pretty much deaf to high frequency phase anyway.

define "high" ?

[dcollins]

lucey wrote on Tue, 25 May 2004 09:14

define "high" ?

Well, let's say deaf above 5kHz.  I think that's what Moore has.  Helmholtz said "phase deaf."   Then Harwood of the Beeb proved we weren't deaf, but not great either.  What is amazing is the brains ability to tell inter-channel differences of like 6 microseconds.  This does not mean you can hear 200kHz, btw.

Maybe with the right wire, though....

DC

[jfrigo]

dcollins wrote on Tue, 25 May 2004 22:11

What is amazing is the brains ability to tell inter-channel differences of like 6 microseconds.  This does not mean you can hear 200kHz, btw. Maybe with the right wire, though....

I try not to stick wires into my brain through my ears, but whatever works for you...

[Techne]

I've been dithering the stereo buss while mixing (i.e. pre-mastering).  Would you ME's rather see an un-dithered track and add your own or is this something i should continue? I'm using POWr currently, but i'd buy the cranesong cd just so i can say i own a piece of their gear  

[TotalSonic]

Techne wrote on Tue, 01 June 2004 13:07

I've been dithering the stereo buss while mixing (i.e. pre-mastering).  Would you ME's rather see an un-dithered track and add your own or is this something i should continue? I'm using POWr currently, but i'd buy the cranesong cd just so i can say i own a piece of their gear

There's no need to add dither when mixing if you're mix is made to a 24bit file or to an analog master - dither is only necessary if you are reducing the word length of the files (i.e. converting from 24bit to 16bit) or digitally summing multiple 16bit tracks with 24bit (or over) internal processing math.

Optimally dithering should just be done once so that there isn't a build up of the low level noise.

If you're mixing digitally best thing to do is to save the mix file as 24bit at the same sample rate you recorded the tracks at and provide that to the ME.

Best regards,
Steve Berson

[bblackwood]

TotalSonic wrote on Tue, 01 June 2004 09:31

Optimally dithering should just be done once so that there isn't a build up of the low level noise.

Sorry, but I think this is bad advice.

Dither using flat (non-noise shaped) dither anytime the wordlength changes. An easy way to think about it is anything you do that changes the sound requires dither, even pulling the fader back 0.001dB.

If you use stage after stage of noise-shaped dither (such as UV-22 or POWr-3) you might experience some nastiness, but you won't notice anything when using flat TPDF whenever you change wordlength.

[Bob Olhsson]

People are missing that when you don't dither, you get distortion that's more audible than the amount of dither necessary to prevent the distortion.

I find undithered reductions to 24 bit can get crunchy at an earlier point in subsequent processing than dithered reductions usually do. If you just listen to the 24bit signal with no additional processing, there isn't much difference. Some people like crunchy while I usually prefer juicy.

I agree that monitoring is far more important than dithering. If the monitoring is really good, the question of dithering takes care of its self. I do think that jitter in a monitor system can make you do things you wouldn't want to do if you heard the same audio without jitter. Jitter can lead you to think there are problems to fix when there really aren't. Modifying the audio to fix a monitoring problem is about the most common mistake I run into.

[TotalSonic]

bblackwood wrote on Tue, 01 June 2004 16:15

TotalSonic wrote on Tue, 01 June 2004 09:31 Optimally dithering should just be done once so that there isn't a build up of the low level noise. Sorry, but I think this is bad advice. Dither using flat (non-noise shaped) dither anytime the wordlength changes. An easy way to think about it is anything you do that changes the sound requires dither, even pulling the fader back 0.001dB. If you use stage after stage of noise-shaped dither (such as UV-22 or POWr-3) you might experience some nastiness, but you won't notice anything when using flat TPDF whenever you change wordlength.

Brad -
Respectfully - you are misinterpreting my comment.  I never stated that you shouldn't dither if changing wordlengths - I agree that if the ME does processing of a 16bit file that has already been dithered using internal processing math that spills over 16bits (i.e. which would usually mean any eq or gain increase or processing with a hardware unit such as the L2)  then additional non-noise shaped dither is appropriate as the final stage after this processing.  

I was speaking of the "optimal" condition if someone is mixing digitally - where there is no wordlength changes made.  You start in 24 bit - and you end in 24 bit.  If you mix entirely in the box - well written modern DAW apps do non-destructive processing/editing using at least 32bit math at every level and round the final track summing result to  24bit once and only once.  If you're sending out via DA to hardware analog processors you send out via 24bit and capture back via AD at 24bit.   And in these cases I simply don't  think additional dither is appropriate.

Best regards,
Steve Berson

[jfrigo]

TotalSonic wrote on Tue, 01 June 2004 13:44

I was speaking of the "optimal" condition if someone is mixing digitally - where there is no wordlength changes made.  You start in 24 bit - and you end in 24 bit.  If you mix entirely in the box - well written modern DAW apps do non-destructive processing/editing using at least 32bit math at every level and round the final track summing result to  24bit once and only once.

Ideal conditions are so hard to come by. Sure, there are conditions when you can get away without dithering, but most people seeking dither advice in on of these forums probably haven't had the experience to distinguish among those close calls yet. It's best if we make it clear to those seeking advice that you should dither for any word length reduction. This includes plug-ins on individual tracks in a mix if they are reducing word length and some mixers in some generations of DAWs. You need to know your DAW and know your plugs. I even dither from 48 bit to 24 bit if I have a choice, just because it can't hurt; it can only help. The extra noise is quite benign, but the distortion from truncation is far more obtrusive.

[TotalSonic]

jfrigo wrote on Tue, 01 June 2004 22:06

I even dither from 48 bit to 24 bit if I have a choice, just because it can't hurt; it can only help. The extra noise is quite benign, but the distortion from truncation is far more obtrusive.

Jay -
Let me ask - WHY?  How is this helping?  What is this noise getting you but an accumulation of noise?  Are we talking about poorly written (and most likely older) DAW apps and badly written plugin algorithms?  Let's not! - these things should be deleted off of peoples drives and replaced with ones that aren't broken - if someone needs one that works I'm a reseller for SAWStudio and can get them a nice discount.  To me good DAW design these days means 100% non-destructive processing - and that processing should never pass a wordlength below 24bit at any point in the processing chain.  

Please correct anything that follows that seems mistaken and please let me know what the logic is that makes it wrong:
Dither in most audio applications is simply noise that helps to mask the fact that in 16bit the least significant bit is still above the noise floor.  In 24bit the least significant bit is below the perceivable noise floor (at least to me in all my listening tests).  So: take a 24bit file.  Make a dsp processing change - have processing math done at either double precision 48bit fixed or 64bit float and return a new result rounded to 24bit.  The least significant bit remains below the perceivable noise floor.  So what is the point of the added noise?  Are you saying the processing math is so poorly written that added noise is needed to get mask the rounding errors even when the word length is staying high enough so that the least significant bit is below the noise floor?  

I'm not doubting that there's a possibility that I'm misunderstanding something here - but I just don't see how if the processing algorithm has a decent enough of resolution so that rounding errors are minimized that dither is necessary at this point.

Best regards,
Steve Berson

[jfrigo]

TotalSonic wrote on Tue, 01 June 2004 22:04

jfrigo wrote on Tue, 01 June 2004 22:06 I even dither from 48 bit to 24 bit if I have a choice, just because it can't hurt; it can only help. The extra noise is quite benign, but the distortion from truncation is far more obtrusive. Jay - Let me ask - WHY?  How is this helping?  What is this noise getting you but an accumulation of noise?  Are we talking about poorly written (and most likely older) DAW apps and badly written plugin algorithms? (SNIP) Dither in most audio applications is simply noise that helps to mask the fact that in 16bit the least significant bit is still above the noise floor.  In 24bit the least significant bit is below the perceivable noise floor (at least to me in all my listening tests).

Uh-oh, I found the problem. Where did this "masking" misunderstanding we see so often come from I wonder? Dither doesn't mask. It actually prevents truncation distorion. There are plenty of places that will give you the technical details, the tech talk page at my website included, so I won't go into it here. If you process at 48 and go back to 24 without dither, you get truncation distortion. Gladly, with a 48 to 24 operation, it's quite low compared to the obvious distortion truncating to 16. A single 48 to 24 truncation you aren't going to hear. A bunch in a row and you might. Technically it absolutely causing truncation distortion (audible or not), the artifacts of which will be several dB greater than the added noise from dither would be, so it's a no brainer to use the dither if it's available.

The only poorly coded software here is the software that does not enable dithering for word length reductions, even reductions to 24 bits. So, while double precision software makes it easier to get away with, especially if you limit it to a couple processes, following technically correct dither practice prevents any possibility of trouble with no negative audible impact, while ignoring it is a crapshoot at 48 (or 64) to 24, and a serious problem from 24 to 16. The noise is far less audible than the distortion. The noise is benign and would take a very large number of processes to become a serious problem, but the resulting distortion from the same number of truncations would be absolutely disasterous.

[TotalSonic]

jfrigo wrote on Wed, 02 June 2004 06:23

The only poorly coded software here is the software that does not enable dithering for word length reductions, even reductions to 24 bits. So, while double precision software makes it easier to get away with, especially if you limit it to a couple processes, following technically correct dither practice prevents any possibility of trouble with no negative audible impact, while ignoring it is a crapshoot at 48 (or 64) to 24, and a serious problem from 24 to 16. The noise is far less audible than the distortion. The noise is benign and would take a very large number of processes to become a serious problem, but the resulting distortion from the same number of truncations would be absolutely disasterous.

hey Jay -
Thanks for clearing this up -
But it seems to me that these types of processes where a plugin for example dithers down to 24bit when returning its final sum are completely transparent to the end user by the design of the application (or if it isn't then the question becomes whether/why the absence is/isn't noticeable by the end user!)  - so I guess going back to the original post - if the word length is not being reduced and the intermediary processing follows good DSP coding and is dithering down before returning its sum to at least 24bit before sending to the next process in line - why should someone go to the bother of adding additional dither on top of this?  And then my final question - what direct listening tests have you done that convinced you of the above point?  I'd like to try them out for myself.

Best regards,
Steve Berson

[jfrigo]

TotalSonic wrote on Tue, 01 June 2004 22:55

hey Jay - Thanks for clearing this up - But it seems to me that these types of processes where a plugin for example dithers down to 24bit when returning its final sum are completely transparent to the end user by the design of the application (or if it isn't then the question becomes whether/why the absence is/isn't noticeable by the end user!)

Sometimes it's transparent to the user, other times it's just left off because they think it unimportant, or that their target market won't know the difference, or don't want to devote any processing cycles or man hours to it. However, it actually is noticeable to many users, even if they don't realize technically what the problem is. Why do so many people complain about "mixing in the box" and why are some plugs better than others? It's not this issue alone, but it certainly contributes.

Quote:

so I guess going back to the original post - if the word length is not being reduced and the intermediary processing follows good DSP coding and is dithering down before returning its sum to at least 24bit before sending to the next process in line - why should someone go to the bother of adding additional dither on top of this?  And then my final question - what direct listening tests have you done that convinced you of the above point?  I'd like to try them out for myself.

If the word length is not reduced, you don't need it, but this is not the case with the majority of things you are likely to do in a DAW. If you move the faders in the mix, it's a gain change, and the word length increases to accommodate the result (provided the DAW has a greater resolution than the file, which it should), and is reduced when you eventually come back to 24 bits. I'm not sure of the specific example you are trying to figure out here. If you are in a true 48 (or 64) bit environment, you don't need to dither from one plug to the next, especially considering that the system probably won't have any additional places to accommodate a larger result from a processed 48 bit word, so it's truncating at 48 anyway. Truncating at 48 I'm really not going to worry about. However, when pro tools HD first came out, it wasn't 48 bits throughout (I think it is now), and it was a good idea to dither to 24 as you came back to the 24 bit mixer from the TDM bus before you handed it off to the next plug or the mix bus. With all those tracks each going through multiple plug ins, it's certainly conceivable that all the resulting truncation distortion could negatively impact your mix.

As for tests, you certainly can use something like Spectrafoo to measure the phenomena and prove it exists, and if you further want to do listening tests, they require a good listening environment, some good tracks (not a bunch of noisy electric guitars), and some time and maybe a friend's assistance to put it together. Based on the concepts in this thread you can devise torture tests if you put your mind to it. It's easier when you happen to have a session in that makes a good example instead of creating one from scratch just for the test, but if you have the time and resources, go for it. Or just torture a good 24 bit stereo piano recording. That may be the simplest first step.

And yes, on several sessions, you may not notice anything depending on the program material, monitoring environment, and if it's only to 24 bits, but that's no reason to follow improper practice if it's simple to avoid. I'm not going to let a ground loop buzz stay around because I'm doing a loud rock project and I don't think anybody will notice. They might, or the next project may not be loud rock. Turning the dither on is easier than tracking down a ground loop, so why not? And there's really no excuse for programmers not to include it.

If your DAW or plug doesn't offer the options where they are needed, then you're out of luck. At that point you decide if you're gaining more than you're losing, and sometimes you definitely are, so don't blow this out of proportion and worry excessively. And if you're unhappy with it, try another DAW or plug. But make no mistake - you'll get truncation distortion if you don't dither, and it's technically not the "correct" way to do it. We do plenty of "technically incorrect" things in making records every day, and that's great, but we should make an informed judgement armed with facts and knowledge when we decide to "break the rules," and not simply do it out of laziness, ignorance, or carelessness.

Technically, the dither is necessary to prevent the distortion. Now that you know, feel free to ignore it when you must, but not because you think it doesn't result in truncation distortion or that the dither is too noisy. Do it because you love the sound of that plug that doesn't offer the dither and because you can't hear any crunchy problems as a result of using it. However, if you have the choice, why risk it?

[lucey]

dcollins wrote on Wed, 26 May 2004 00:11

Well, let's say deaf above 5kHz.  I think that's what Moore has.  Helmholtz said "phase deaf."   Then Harwood of the Beeb proved we weren't deaf, but not great either.  What is amazing is the brains ability to tell inter-channel differences of like 6 microseconds.  This does not mean you can hear 200kHz, btw. Maybe with the right wire, though....

very funny



i'm still thinking we can feel the out of phaseness above 5k, if not actually hear it.

would the microseconds be sufficient to sense it even if the ear cannot 'hear' it, technically.



basically, was this a hearing test or a perception test ... with "5k" as a result ?