Dubstepforum.com

While I think I understand the principle more or less, I've never really actively used it too much.

I'm bouncing down to stems in Reaper right now and there's the "dither" box... never touched it.

So do you dither? Any advice from M.E. types? Any reason I should[n't]?

I always dither when converting a 24 bit file to a 16 bit. I'm not the most qualified guy to explain why though.

I would like to know more about the differences with the algorithms, and what happens when you dither with a rectangle vs. triangle for instance.

I don't dither anymore. I cannot tell the fucking difference!

So I had to read what dither means. I wish I did before normalizing a bunch of wicked samples earlier today.
Erf anyway... I probably can't tell the difference either

And that, my friends, is why I'm skeptical whenever people say they can hear the difference between 24 and 16 bit on mastered material.

static_cast wrote:And that, my friends, is why I'm skeptical whenever people say they can hear the difference between 24 and 16 bit on mastered material.

I'm not even sure I'd pass a 192 kbps vs wav blind test

I tend to agree. I think perhaps it's something where in cases where extreme quality & fidelity is required - 96 > 24 mastering on symphonic stuff, perhaps.

With the shit setup I'm working with though? I kinda doubt it makes any noticeable diff.

Well you want to mixdown your premaster to 24 bit because of the processing in the mastering stages. You could theoretically start to hear an accumulative effect. That said I know for sure that 320's have been used for masters in cases where all other files are lost. And it's put out at the usual places where the public is non suspecting. Just sayin.

Basically, when you convert from 24 bit to 16 bit the 8 'least significant bits' (a.k.a. the bits with the least amount of data on it) is removed (which is why the headroom tends to be a problem when downsampling) and something called 'quantisation error' can occur, which presents itself in the form of nasty artifacts; by reducing bit rate, errors can occur when a sample exists on a value that doesn't exist in the new bit rate (for the sake of argument i'm going to give the example of 4.5; it has to change to either 4 or 5 upon resample, both of which are wrong).

Dither is essentially white noise intentionally placed on top of the track to make the error (and resulting artifacts) less noticeable, and to help make it less audible the noise can be sculpted to avoid certain frequency ranges. As a rule, if you look at a dithered piece of audio through a frequency analyser (this is easier when you do it when the track is silent or at it's quietest) you'll see that there's a 'blanket' of noise throughout most frequencies with big spikes in noise in the 0-40Hz and 18-20KHz ranges that's dither!

Hope this sheds some light on dither peeps :3


Dub Fiend

abZ wrote:I don't dither anymore. I cannot tell the fucking difference!

Same.

Dub Fiend wrote:Basically, when you convert from 24 bit to 16 bit the 8 'least significant bits' (a.k.a. the bits with the least amount of data on it) is removed (which is why the headroom tends to be a problem when downsampling) and something called 'quantisation error' can occur, which presents itself in the form of nasty artifacts; by reducing bit rate, errors can occur when a sample exists on a value that doesn't exist in the new bit rate (for the sake of argument i'm going to give the example of 4.5; it has to change to either 4 or 5 upon resample, both of which are wrong).

Dither is essentially white noise intentionally placed on top of the track to make the error (and resulting artifacts) less noticeable, and to help make it less audible the noise can be sculpted to avoid certain frequency ranges. As a rule, if you look at a dithered piece of audio through a frequency analyser (this is easier when you do it when the track is silent or at it's quietest) you'll see that there's a 'blanket' of noise throughout most frequencies with big spikes in noise in the 0-40Hz and 18-20KHz ranges that's dither!

Hope this sheds some light on dither peeps :3


Dub Fiend

Thought dither was just white noise, one bit loud, and the "shape" was actually just the shape of the probability distribution (Gaussian, triangular etc)?

Either way, if you can't hear the dithering noise, you can't hear the quantization error either, and therefore dithering makes no audible difference. (But as with many other things in the audio world, that doesn't necessarily mean you shouldn't do it).

Came across an unusual method of dithering , had an ME that ran the mastered 24/48 audio through a manley pre amp and recorded it back in at 16/44.1.

danoldboy wrote:Came across an unusual method of dithering , had an ME that ran the mastered 24/48 audio through a manley pre amp and recorded it back in at 16/44.1.

Oh yeah, that's the other thing. If you've recorded any hardware and the noise floor of your gear is over -96dB then in theory there's not much point dithering either. Ditto if you've bounced to tape.

Dub Fiend wrote:Basically, when you convert from 24 bit to 16 bit the 8 'least significant bits' (a.k.a. the bits with the least amount of data on it) is removed (which is why the headroom tends to be a problem when downsampling) and something called 'quantisation error' can occur, which presents itself in the form of nasty artifacts; by reducing bit rate, errors can occur when a sample exists on a value that doesn't exist in the new bit rate (for the sake of argument i'm going to give the example of 4.5; it has to change to either 4 or 5 upon resample, both of which are wrong).

Dither is essentially white noise intentionally placed on top of the track to make the error (and resulting artifacts) less noticeable, and to help make it less audible the noise can be sculpted to avoid certain frequency ranges. As a rule, if you look at a dithered piece of audio through a frequency analyser (this is easier when you do it when the track is silent or at it's quietest) you'll see that there's a 'blanket' of noise throughout most frequencies with big spikes in noise in the 0-40Hz and 18-20KHz ranges that's dither!

Hope this sheds some light on dither peeps :3


Dub Fiend

Is this why when i upload 320's to soundcloud I get these popping type artifacts on playback?

naus wrote:

Dub Fiend wrote:Basically, when you convert from 24 bit to 16 bit the 8 'least significant bits' (a.k.a. the bits with the least amount of data on it) is removed (which is why the headroom tends to be a problem when downsampling) and something called 'quantisation error' can occur, which presents itself in the form of nasty artifacts; by reducing bit rate, errors can occur when a sample exists on a value that doesn't exist in the new bit rate (for the sake of argument i'm going to give the example of 4.5; it has to change to either 4 or 5 upon resample, both of which are wrong).

Dither is essentially white noise intentionally placed on top of the track to make the error (and resulting artifacts) less noticeable, and to help make it less audible the noise can be sculpted to avoid certain frequency ranges. As a rule, if you look at a dithered piece of audio through a frequency analyser (this is easier when you do it when the track is silent or at it's quietest) you'll see that there's a 'blanket' of noise throughout most frequencies with big spikes in noise in the 0-40Hz and 18-20KHz ranges that's dither!

Hope this sheds some light on dither peeps :3


Dub Fiend

Is this why when i upload 320's to soundcloud I get these popping type artifacts on playback?

doubtful

naus wrote:

Dub Fiend wrote:Basically, when you convert from 24 bit to 16 bit the 8 'least significant bits' (a.k.a. the bits with the least amount of data on it) is removed (which is why the headroom tends to be a problem when downsampling) and something called 'quantisation error' can occur, which presents itself in the form of nasty artifacts; by reducing bit rate, errors can occur when a sample exists on a value that doesn't exist in the new bit rate (for the sake of argument i'm going to give the example of 4.5; it has to change to either 4 or 5 upon resample, both of which are wrong).

Dither is essentially white noise intentionally placed on top of the track to make the error (and resulting artifacts) less noticeable, and to help make it less audible the noise can be sculpted to avoid certain frequency ranges. As a rule, if you look at a dithered piece of audio through a frequency analyser (this is easier when you do it when the track is silent or at it's quietest) you'll see that there's a 'blanket' of noise throughout most frequencies with big spikes in noise in the 0-40Hz and 18-20KHz ranges that's dither!

Hope this sheds some light on dither peeps :3


Dub Fiend

Is this why when i upload 320's to soundcloud I get these popping type artifacts on playback?

Do you have an example? Id suspectzero crossings, I never notice if there in my tunes until I upload...

static_cast wrote: Oh yeah, that's the other thing. If you've recorded any hardware and the noise floor of your gear is over -96dB then in theory there's not much point dithering either. Ditto if you've bounced to tape.

Not necessarily true - you can hear and discern information below the noise floor. Happens all the time in real life (think about it ).

It's important to stress that dither isn't just white noise. Broadband noise yes, but as mentioned, statistically shaped.

Also not entirely true that you see peaks in the low and high end - that depends on what noise shaping you are using. For almost all material, TPDF (ie 'flat') is good enough. If there are very quiet/silent parts then shaping might be the way to go, though it can make things brittle if you ask me (at least, with heavier noise shaping).

On super-maximised stuff of course it is debateable whether it makes such an audible difference, but then almost every tune has at least one bit where it breaks down/goes a bit quieter - these are the parts where you should listen for it if you're that interested.

From my point of view, I spend about half a second considering dither on each track I work on. When a track has 8dB too much 50Hz, there are bigger issues to be considering! It has to be said that while I don't always directly hear the effect unless doing listening tests for it, on occasion I have found shaped dithers to mess with the high end I just spent ages getting right. Now I plump for TPDF on 99% of material, safe in knowing I am not adding any distortion or messing with the frequency balance.


The simplest way to consider it is that going from 24 > 16 bit is bit-crushing, plain and simple. We all know what that sounds like, and even if it is orders of magnitude lower, it's still there. Dither reduces the distortion associated with that. Whether you can hear it or not NOW (cos you may in future), if all it means is ticking a tick box, wouldn't you rather know you're avoiding it?

macc wrote:

static_cast wrote: Oh yeah, that's the other thing. If you've recorded any hardware and the noise floor of your gear is over -96dB then in theory there's not much point dithering either. Ditto if you've bounced to tape.

Not necessarily true - you can hear and discern information below the noise floor. Happens all the time in real life (think about it ).

True, but my understanding was that the dither just makes the quantisation error random, rather than directly correlated to the signal. Don't you only need 1 effective bit of additive noise in order to do this?

It's important to stress that dither isn't just white noise. Broadband noise yes, but as mentioned, statistically shaped.

Also not entirely true that you see peaks in the low and high end - that depends on what noise shaping you are using. For almost all material, TPDF (ie 'flat') is good enough. If there are very quiet/silent parts then shaping might be the way to go, though it can make things brittle if you ask me (at least, with heavier noise shaping).

On super-maximised stuff of course it is debateable whether it makes such an audible difference, but then almost every tune has at least one bit where it breaks down/goes a bit quieter - these are the parts where you should listen for it if you're that interested.

From my point of view, I spend about half a second considering dither on each track I work on. When a track has 8dB too much 50Hz, there are bigger issues to be considering! It has to be said that while I don't always directly hear the effect unless doing listening tests for it, on occasion I have found shaped dithers to mess with the high end I just spent ages getting right. Now I plump for TPDF on 99% of material, safe in knowing I am not adding any distortion or messing with the frequency balance.


The simplest way to consider it is that going from 24 > 16 bit is bit-crushing, plain and simple. We all know what that sounds like, and even if it is orders of magnitude lower, it's still there. Dither reduces the distortion associated with that. Whether you can hear it or not NOW (cos you may in future), if all it means is ticking a tick box, wouldn't you rather know you're avoiding it?

I'm totally with you, I was just being finnickity for the sake of it. It is just a tickbox after all; I never said you shouldn't...

macc wrote:Also not entirely true that you see peaks in the low and high end - that depends on what noise shaping you are using.

Should've put that really, I do know about noise shaping hehe

I get what naus means about artifacts on SC, but I think that's more of a sample rate problem more than anything; I believe they playback at 128kbps, and speaking from experience I can hear an audible 'buzz' when my sub bass plays so I know it's not just in your head mate Lowering the sample rate reduces the amount of frequencies available, so I know that (at least in my case) sample rate is the issue, as the extremely low and high frequencies are sheared off

This whole "if you can't hear the dithering noise, you can't hear the quantization error either" thing isn't true though, I mean it's not that audible but if I had the choice of not using dither and having buzzing noises on my reverb tails or just turning it on, I'd pick the latter every time.

Oh, I found a video that explains it on YT too and he gives you examples of what quantisation error sounds like



The guy sounds like a bit of a douche, but it's cool


Dub Fiend