|
What the hell is dithering ? (pg. 3)
|
View this Thread in Original format
| Derivative |
| quote: | | Are the loops/samples/drum sounds availabe on sample cds/dvds and on the net dithered or undithered? |
the loops and hits you get on sample cds and on the net are all undithered. at least, those ones that are professionally made. cant speak for some of the cowboy sample packs that get uploaded though.
| quote: | So, in the end, is it always best to dither when converting from 24bit to 16bit?
I mean, will a 'regular' listener possibly hear the difference from a sample that has been dithered and one that hasn't been? Or is this just some sort of 'beauty trick' for engineers? |
it doesnt strictly matter. you would use dither whenever you are downsizing the bitrate of an audio recording. the more bits you lose though, the more quantisation noise is evident.
do you notice a big difference between your 24 bit internal mixes and the rendered 16 bit result? i notice differences in mine, even on pretty cheap hifi speakers. but thy pretty much hit the nail on the head though. if you listen out for it you will notice it, although its very very subtle. i have to do A/B comparisons to notice anything. but to the average joe who will be listening to your track on his ipod, and who doesnt even know what quantisation noise is - they wont have a clue what the difference is. of course, the average joe wouldnt have access to your 24 bit pre masters, so consider the fact that you dont dither to be icone's dirty little secret. we know your secret icone...bwahaha.
i guess you could call it a beauty trick for engineers. if you are like me and are still fuking up in the mixdown then theres absolutely no point in thinking about dither. i can get more noticeable results from improving my production at this point. since production is an ongoing learning process for alot of people, it is in my humble opinion that this is best left to a mastering engineer who will more than likely have an awesome set of ears, and an even more awesome set of monitors to fine tune your work.
but dither i think is just one of those final, final touches. the tiniest, final layer of polish to try and make a record as absolutely perfect as it can possibly be, given the limits of the medium it has to be recorded to.
| quote: | | So let's all be clear on this - dithering is not the process of adding noise, that just happens to be a side effect of certain quantization methods. |
diginut. i know you like disagreeing with me. i know you luv it in fact, and you live to see my face, bruised and beaten in the gutter, KO'ed by a digi fist :grin: but dither *is* the process of adding noise to a recording. those algorythms you mentioned i thought had already been cleared up by thy as dithering + noise shaping via psychoacoustic principles to make dithering noise less evident.
you can spin it whatever way you like, and quote however many jargon terms. but when you dither, you are adding random noise to a recording. period. |
|
|
| DigiNut |
| quote: | Originally posted by Derivative
diginut. i know you like disagreeing with me. i know you luv it in fact, and you live to see my face, bruised and beaten in the gutter, KO'ed by a digi fist :grin: but dither *is* the process of adding noise to a recording. those algorythms you mentioned i thought had already been cleared up by thy as dithering + noise shaping via psychoacoustic principles to make dithering noise less evident. |
Haha. I'm just going to repeat what I said before though - dithering is not the process of adding noise, and that's the wrong way of looking at it. Noise is inevitable when lowering the bitrate - dithering is really the process of trying to eliminate as much of that noise as possible. |
|
|
| Icone |
| Don't get me wrong, I do use dithering when doing the end-mastering in T-Racks. I know it is actually 'good' for your track when finalising and downgrading the bitsize. Though, I just didn't know the specifics (don't know it that's really necessary) ;) |
|
|
| harriz |
| quote: | Originally posted by Derivative
|
Its masking unpleasent noise from the lowering a bitrate with noise that is more pleasent,random and less harsh
from what I understand here so I guess this process could be viewed as removing noise .
Does normalising te recording add noise or not?
Would normalising a recording ruin the quality of a track if it was applied more than onse ? |
|
|
| Atlantis-AR |
| quote: | Originally posted by Storyteller
Any decent sequencer works 32bit floating point internally when it comes to sound processing. in the end sound wil be translated back to the sound properties of the track settings. Any studio god-like guy taking himslef seriously works in the bitrate/hz range which is most likely to be the final processing rate.
So if you're producing for a dvd it's likely your whole project would be 96khz/24bit, in this case you don't need to dither at all, overcoming all these seemingly notable (by uber leet pro's :D) problems..
So if you're working for a cd this would mean you'd work on 44.1khz 16 bit the entire project. |
You know, this isn't true at all. Process a master, let alone a mix, at 88.2 kHz or 96 kHz, after which resampling it back to 44.1 kHz, and your life will never be the same. ;)
And you say so yourself that most software processes internally at 32 bit floating point precision (which I believe is true), but then what happens when you render your file at a resolution of 16 bits (abiding by your approach of working at CD (44/16) resolution)? Are the extra bits truncated, or is dithering added before downsampling? I would surely hope it's dithering, the option for which you will find under Configs->Audio in Renoise. |
|
|
| Atlantis-AR |
| quote: | Originally posted by Derivative
the loops and hits you get on sample cds and on the net are all undithered. at least, those ones that are professionally made. cant speak for some of the cowboy sample packs that get uploaded though. |
I always thought the opposite. Can you explain to me what happens then when someone records a bunch of samples at 44,100 Hz, 24 bit, and intends on releasing them at a resolution of 16 bits? Shouldn't you be applying dithering then, but just using a more moderate amount of noise shaping? |
|
|
| DigiNut |
| quote: | Originally posted by Atlantis-AR
And you say so yourself that most software processes internally at 32 bit floating point precision (which I believe is true), but then what happens when you render your file at a resolution of 16 bits (abiding by your approach of working at CD (44/16) resolution)? Are the extra bits truncated, or is dithering added before downsampling? I would surely hope it's dithering, the option for which you will find under Configs->Audio in Renoise. |
In Cubase I just stick a UV22-HR on the last slot of the master channel and render it as 16 bits. For offline mastering (which I don't usually do), there's always the option of rendering at 32 bits and using the UV22-HR or Waves L1/L2/L3 in the dither slot in Wavelab. |
|
|
| Atlantis-AR |
| quote: | Originally posted by DigiNut
In Cubase I just stick a UV22-HR on the last slot of the master channel and render it as 16 bits. For offline mastering (which I don't usually do), there's always the option of rendering at 32 bits and using the UV22-HR or Waves L1/L2/L3 in the dither slot in Wavelab. |
Yeah, that's what I would do. I was just saying to Storyteller that Renoise already dithers by default when you render down to 16 bits. When you render as 32 bit float (in Renoise), I'm quite positive the dither is bypassed either way, and you can then apply it when you master in an external program. Alternatively, you could still bypass the dither in Renoise, and stick an L3 on the master channel in Renoise to add dithering (still rendering as a 16 bit file). I would imagine this produces superior results, though no different from rendering as 32 bit float and dithering afterwards. |
|
|
| Storyteller |
| quote: | Originally posted by Atlantis-AR
You know, this isn't true at all. Process a master, let alone a mix, at 88.2 kHz or 96 kHz, after which resampling it back to 44.1 kHz, and your life will never be the same. ;) |
Strange how the quality of sound can differ when the Hz and bitrate in the end is the same. Never tried it and never had any problems about it :) Also I've seen a poll in a music magazine, more than half of professional audio studio's worked on a different quality setting adapted to the final medium. So I know it's true for a lot of dutch studio engineers :), maybe in case of mastering a song this is quite a bit different.
| quote: | | And you say so yourself that most software processes internally at 32 bit floating point precision (which I believe is true), but then what happens when you render your file at a resolution of 16 bits (abiding by your approach of working at CD (44/16) resolution)? Are the extra bits truncated, or is dithering added before downsampling? I would surely hope it's dithering, the option for which you will find under Configs->Audio in Renoise. |
Good one, I really have no clue but I'm very interested in what the outcome is for other software packages, what happens when the output setting is 16bit and nobody ditheres the sound just before. I believe dithering is on in my case :) |
|
|
| Atlantis-AR |
| quote: | Originally posted by Storyteller
Strange how the quality of sound can differ when the Hz and bitrate in the end is the same. Never tried it and never had any problems about it :) |
Re sample rate, there's a thing known as "aliasing distortion", which is, one, beyond the scope of this topic, and, two, still a little outside my own knowledge to explain it in the best light, so Google is where you'll want to be if you're interested. Just know that processing at a higher sample rate, after which downsampling to the prefered rate will yield better results than processing at that lower sample rate to begin with.
And on that note, that's why the quality differs so much when the bit rate isn't the same either, as you're essentially processing the sound data (adding an effect, resampling the sound, etc.) rather than performing a direct copy. If you have a 16 bit sample, process it with an DSP working at a resolution of 16 bits also, and apply this transformation to the sample, you're going to get inferior results than processing the sound at a higher bit depth, such as 32 bit floating point. The extra bits allow for more intermediate sample values, giving off a more natural sound, free from quantisation noise.
So that's why processing at a higher bit depth is so important, after which you should apply dithering to keep the amount of additional quantisation noise added to a minimum.
| quote: | | Also I've seen a poll in a music magazine, more than half of professional audio studio's worked on a different quality setting adapted to the final medium. So I know it's true for a lot of dutch studio engineers :), maybe in case of mastering a song this is quite a bit different. |
So if I understand you right, you're saying that professional studios do work at higher sample and bit rates, such as 96 kHz, 24 bit? Well, there you go. ;)
And, yes, it does help greatly during mastering, in retaining as much information as possible, but that's no reason the mix can't be processed at 96 kHz as well. It requires a hell of a lot more processing power, but in the near future, I think (or hope at least) 96 kHz will become a new standard amongst project studio, at least during processing.
| quote: | | Good one, I really have no clue but I'm very interested in what the outcome is for other software packages, what happens when the output setting is 16bit and nobody ditheres the sound just before. I believe dithering is on in my case :) |
I'm quite convinced that every good software package applies dithering.
Anyway, someone please correct my if I'm wrong anywhere, as I've never been as strong on the technical and mathematical side of things than on the practical side of mastering. |
|
|
| Dj Thy |
| quote: | Originally posted by harriz
Its masking unpleasent noise from the lowering a bitrate with noise that is more pleasent,random and less harsh
from what I understand here so I guess this process could be viewed as removing noise .
Does normalising te recording add noise or not?
Would normalising a recording ruin the quality of a track if it was applied more than onse ? |
If you look at it the strict way, any digital process except a pure cut/copy/paste will deteriorate the signal in some way. Even a simple gain change (and face it, normalise is just a gain).
An example in simple terms (it's a metaphor I like using) :
Say you have 7.5$, with that resolution (one character after the decimal, .5). You want to increase the amount of money by 1.243 (this would be the gain you apply on your signal). 7.5*1.243=9.3225. But you don't have that big a resolution. So you'll end up with 9.3$ Basically, you miss 0.0225$. Same goes for a negative gain. And gain is only a pretty simple process...
Doesn't seem like much, but in audio, that microinformation is what gives depth and space to the sound.
Indeed, if you work in higher resolution, the errors induced by processes will be smaller. You'll probably ask what's the point if you decrease the resolution in the end. You'll use the extra info anyway.
Well, that's right, but with decent dithering algo's, you'll still retain some of it.
If stuff like this interests you guys, I can only recommend the book Mastering Audio by Bob Katz, which is an excellent read for all the people that want to know what's going when you deal with sound. I strongly recommend it to anyone, not only to aspiring mastering engineers. |
|
|
| harriz |
| quote: | Originally posted by Dj Thy
If you look at it the strict way, any digital process except a pure cut/copy/paste will deteriorate the signal in some way. Even a simple gain change (and face it, normalise is just a gain).
An example in simple terms (it's a metaphor I like using) :
Say you have 7.5$, with that resolution (one character after the decimal, .5). You want to increase the amount of money by 1.243 (this would be the gain you apply on your signal). 7.5*1.243=9.3225. But you don't have that big a resolution. So you'll end up with 9.3$ Basically, you miss 0.0225$. Same goes for a negative gain. And gain is only a pretty simple process...
Doesn't seem like much, but in audio, that microinformation is what gives depth and space to the sound.
Indeed, if you work in higher resolution, the errors induced by processes will be smaller. You'll probably ask what's the point if you decrease the resolution in the end. You'll use the extra info anyway.
Well, that's right, but with decent dithering algo's, you'll still retain some of it.
If stuff like this interests you guys, I can only recommend the book Mastering Audio by Bob Katz, which is an excellent read for all the people that want to know what's going when you deal with sound. I strongly recommend it to anyone, not only to aspiring mastering engineers. |
Thanks for the recommendation Dj thy ,its not very expensive, I will order this online.
Do you think its comprehensive enough for a beginner to understand ? |
|
|
|
|
