|
|
Post by Bob Olhsson on Jul 6, 2014 15:40:35 GMT -6
Dither makes the gating of the bottom bits being chopped off sound like random noise instead of chattering along with the program material. Every single bit of signal processing we do expands the bit rate and requires dithering when the bits get reduced. Dither doesn't just cover up the distortion, it prevents it.
I understand the requirement of dither was determined at Bell Labs during the 1950s. Unfortunately the digital audio wheel has been reinvented a number of times. The most neutral dither I've heard is Sonnox TPDF which was developed by Paul Frindle for the Sony Oxford console using extensive double blind listening tests.
|
|
|
|
Post by popmann on Jul 7, 2014 7:37:21 GMT -6
So, couple things to note: the 32bit thing is EASY to wrap your head around. It's the size of a pipe. Your ADC only does 24bit (and only ever will-see physics and component self noise)....so your FILES are 24bit--that's all you captured. But, that's neither hear nor there (pun intended)....in terms of the size of the pipe in the digital mixer. I worked for years on a 56bit fixed digital mixer. More on that in a second....point is you can ALWAYS throw less bits down greater bits smaller data down a bigger pipe. Right? So, the STREAM of digital audio that a 32bit float software mixer outputs IS 32bitfloat--whether your app "supports" capturing or storing in that file format. So, you are, by way of saving a 24bit mix file reducing the bit depth...but, if you want to get into a mindfuck--you've always been listening to a truncated stream via your 24bit DAC, so....
So, the answer is that in THEORY, it's safest to apply dither any time there's a bit reduction. This might be on a double precision plug in...the output of your mixer....theoretically, every hardware insert, as you're taking a 32bit stream and feeding it to a 24 bit DAC....but, I continue to use theoretically, because I've never heard the magic of dither. My Akai workstation working at 56bit never had ANY dither options. Does that mean it just did it automagically? Maybe--it certainly did a lot that way....or did it simply chop the stream? Anyway--I've done a handful of with/without (albeit always going all the way to 16/44 CD) and I could never tell any repeatable difference. But, before he who shall not be named jumps in to say how we finally agree...we don't--I understand the THEORY of how it helps...therefore, I do it. Because why not? A tiny amount of noise? Ha. My amps and Leslie and mics-absurd analog modeling plug ins that have way more hiss than the actual units....I'm not going to approach the SNR where dither would be terribly audible.
I capturethe 32bit float output (because I can)--and open the UV22HR, optimize for 24bit-spit it out, seconds later, optimize for 16bit, spit it out....so, within one minute of printing a mix, I have a (proprietary) 32bit float file, 24bit, and 16bit--each with their optimized dither levels.
Wass that too long? Should I have simply said "listen to Bob"? Maybe. This falls under the rule for me of "I don't hear it....but, I do it because theory says to....AND there's no downside"--maybe I do hear it. It's simply not worth the time for me to ABX and compare....because what's the price? some tiny DB of noise? Half a Mhz of CPU? I'll trust people who have spent the time.
I will say WavesIDR's "ultra" mode should be never EVER fucking used. Chased that "where's the scratchy sound coming from" in a mix for an hour one day....only to find I'd set it there on some plug in thinking "yeah, dude....ULTRA....that's what I want" 9I guess, because it's not a default)....absolutely audible....absolutely garbage. I've never heard such with any other or Waves IDR left at whatever default.
|
|
|
|
Post by tonycamphd on Jul 7, 2014 8:43:56 GMT -6
Dither prevents low level distortion. The only time to not dither is when copying audio unchanged.
A tenth of a dB. level change expands bit depth beyond what can be handed to the next process or written to a file. Hence it requires dither to prevent edgy "digital sounding" distortion. It can be hard to hear with a DAW until after you write the file because many plug-ins let the DAW write the truncation but the monitoring is still 24 bit. The other thing about the distortion is that it accumulates twice as fast with every generation than the noise from dithering does. If the audio has ever been truncated, the benefits of 24 bit dither become much harder to hear. This has led many people to think it isn't important. If not, the dimension and depth both go away when you turn off the dither. On top of that signal processing starts sounding edgy.
Properly engineered dither is less audible than the distortion caused by not dithering. Dither being an option is lazy programming. We ought to just be able to select an output format and never need to think about dither. It's just as stupid as having a separate record bias button would be on an analog tape machine. Thanx Sir Bob, i want to make sure i'm understanding this correctly, you are saying that if i perform any changes(mixing) to a 96/24bit session file, i should bounce with dither even if i'm bouncing a same 96/24bit file?? |
|
|
|
Post by tonycamphd on Jul 7, 2014 8:48:42 GMT -6
So, couple things to note: the 32bit thing is EASY to wrap your head around. It's the size of a pipe. Your ADC only does 24bit (and only ever will-see physics and component self noise)....so your FILES are 24bit--that's all you captured. But, that's neither hear nor there (pun intended)....in terms of the size of the pipe in the digital mixer. I worked for years on a 56bit fixed digital mixer. More on that in a second....point is you can ALWAYS throw less bits down greater bits smaller data down a bigger pipe. Right? So, the STREAM of digital audio that a 32bit float software mixer outputs IS 32bitfloat--whether your app "supports" capturing or storing in that file format. So, you are, by way of saving a 24bit mix file reducing the bit depth...but, if you want to get into a mindfuck--you've always been listening to a truncated stream via your 24bit DAC, so.... So, the answer is that in THEORY, it's safest to apply dither any time there's a bit reduction. This might be on a double precision plug in...the output of your mixer....theoretically, every hardware insert, as you're taking a 32bit stream and feeding it to a 24 bit DAC....but, I continue to use theoretically, because I've never heard the magic of dither. My Akai workstation working at 56bit never had ANY dither options. Does that mean it just did it automagically? Maybe--it certainly did a lot that way....or did it simply chop the stream? Anyway--I've done a handful of with/without (albeit always going all the way to 16/44 CD) and I could never tell any repeatable difference. But, before he who shall not be named jumps in to say how we finally agree...we don't--I understand the THEORY of how it helps...therefore, I do it. Because why not? A tiny amount of noise? Ha. My amps and Leslie and mics-absurd analog modeling plug ins that have way more hiss than the actual units....I'm not going to approach the SNR where dither would be terribly audible. I capturethe 32bit float output (because I can)--and open the UV22HR, optimize for 24bit-spit it out, seconds later, optimize for 16bit, spit it out....so, within one minute of printing a mix, I have a (proprietary) 32bit float file, 24bit, and 16bit--each with their optimized dither levels. Wass that too long? Should I have simply said "listen to Bob"? Maybe. This falls under the rule for me of "I don't hear it....but, I do it because theory says to....AND there's no downside"--maybe I do hear it. It's simply not worth the time for me to ABX and compare.... because what's the price? some tiny DB of noise? Half a Mhz of CPU? I'll trust people who have spent the time.
I will say WavesIDR's "ultra" mode should be never EVER fucking used. Chased that "where's the scratchy sound coming from" in a mix for an hour one day....only to find I'd set it there on some plug in thinking "yeah, dude....ULTRA....that's what I want" 9I guess, because it's not a default)....absolutely audible....absolutely garbage. I've never heard such with any other or Waves IDR left at whatever default. Thanx Pop, could this be a sum of small things example? 40 channels of tiny db's of chatter adding up to... chatter? Maybe this is the "digititus" i hear, that jim williams coined? |
|
|
|
Post by Bob Olhsson on Jul 7, 2014 11:26:44 GMT -6
That's right, all digital signal processing needs to be dithered.
You may not hear one generation of truncation distortion but after several the sound becomes brittle and crunchy aka "digititus." I've sent a number of mastering projects back for another bounce using the dithered mixer or 24 bit dither in Pro Tools. I've never once been wrong in thinking it hadn't been dithered. I'm sure I've also mastered plenty of projects where it wasn't dithered but when things start sounding crunchy, dither has just about always turned out to be the fix.
Dither subtracts distortion, it is NOT just adding noise which is a common misconception. Dither also sounds like noise but noise does not necessarily function as dither. The spectrum and level are both critical. There have also been plenty of broken dither routines in software over the years which has added to the confusion.
|
|
|
|
Post by littlesicily on Jul 7, 2014 16:18:26 GMT -6
|
|
|
|
Post by lolo on Jul 7, 2014 22:49:38 GMT -6
Yeah great thread. Thanks for the info popmann and Bob Olhsson
Just to clarify one more thing. If im processing individual tracks with plugins and hardware inserts and I would like to bounce/print to save some processing power. Is it still recommended to insert a dither plug on the last insert before I print?
Thanks |
|
|
|
Post by jeromemason on Jul 7, 2014 23:13:39 GMT -6
My issue I think was using the Ultra on Waves, I was getting some terrible scratchy crud that didn't exist in my full res print, but put it on a CD or MP3, and I'm suddenly in amateur hour.
I'm giving the PSP x-dither a go, and I'm really liking all of it's options. It's not expensive either, but it's not cheap either. I'd love to find a good dither plugin I rely upon so that when I print and send off I don't have to even think about any of that shit. This thread rocks, and it has helped me find a problem that I've been having for a while now, since I've started having to jack the volume up on Ref mixes.
Thanks Bob, and POP, your input is invaluable in regards to this, and many folks owe both of you many thanks for taking time to explain this to a bunch of folks that probably didn't have the first damn clue what dither was and when to use it.
I believe I'll be safe as long as when I print a file, I print it with a 24 bit, and a 16 bit dither. Then I can simply export in whatever format they want, .wav, mp3, AAC, whatever, and I should be safe. Yes?
|
|
|
|
Post by tonycamphd on Jul 7, 2014 23:33:03 GMT -6
Yeah great thread. Thanks for the info popmann and Bob Olhsson
Just to clarify one more thing. If im processing individual tracks with plugins and hardware inserts and I would like to bounce/print to save some processing power. Is it still recommended to insert a dither plug on the last insert before I print?
Thanks this is a good question i'd like to know the answer to as well. My sitch is a little different, i'm using plug filtering and limited plug effects, then coming out single channels into my console, it gets voltage based summing from that point, should i be using a dithering plug on each ITB track that has any automation/ plugs instantiated before leaving the digital domain? yeah, Bob O and Pop, good stuff guys! thanx |
|
|
|
Post by Bob Olhsson on Jul 8, 2014 6:38:25 GMT -6
Yes! Try it and see.
|
|
|
|
Post by littlesicily on Jul 9, 2014 14:20:31 GMT -6
So while recording at 24bits in PT, when I go OTB for processing I should dither? And then when I print my mix ITB I should dither again?
Of the 3 Dither plus I have: Digi Dither, POW-r Dither and Waves IDR… none of them quantize to 24 bits. Only 20 or 16 bits. How do I stay at 24 bits? Are there other Dither plugs u guys know of and like?
|
|
|
|
Post by henge on Jul 9, 2014 14:54:13 GMT -6
Dither prevents low level distortion. The only time to not dither is when copying audio unchanged.
A tenth of a dB. level change expands bit depth beyond what can be handed to the next process or written to a file. Hence it requires dither to prevent edgy "digital sounding" distortion. It can be hard to hear with a DAW until after you write the file because many plug-ins let the DAW write the truncation but the monitoring is still 24 bit. The other thing about the distortion is that it accumulates twice as fast with every generation than the noise from dithering does. If the audio has ever been truncated, the benefits of 24 bit dither become much harder to hear. This has led many people to think it isn't important. If not, the dimension and depth both go away when you turn off the dither. On top of that signal processing starts sounding edgy.
Properly engineered dither is less audible than the distortion caused by not dithering. Dither being an option is lazy programming. We ought to just be able to select an output format and never need to think about dither. It's just as stupid as having a separate record bias button would be on an analog tape machine. So ANYTIME we bounce a track down after processing we should dither right? |
|
|
|
Post by Bob Olhsson on Jul 10, 2014 10:07:46 GMT -6
Every time as part of the processing! It's choosing between distortion or less audible noise. if there's no easy way, bounce to a floating point file.
Because dither requires extra DSP power, developers made it optional so they could advertise more features.
|
|
|
|
Post by littlesicily on Jul 10, 2014 13:47:59 GMT -6
Every time as part of the processing! It's choosing between distortion or less audible noise. if there's no easy way, bounce to a floating point file. I'm trying to figure out a practical way to do this in Pro Tools in a fast-paced workflow. For example, I'll often add a fade to the start and end of a clip and then consolidate it to a new file. Bouncing in realtime is not a good solution and I don't have an audio suite plug that dithers to 24bits. Any ideas anyone? |
|
|
|
Post by mulmany on Jul 10, 2014 13:56:53 GMT -6
If you are using the Fade function in PT then it has a check box for adding dither.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jul 10, 2014 14:10:37 GMT -6
Siding with Henge and Johnkenn. I usually send clients just an MP3 from the session export dialog, it's a pretty hi res MP3 that I don't think I'd pick out against the master file. I just don't hear it.
|
|
|
|
Post by littlesicily on Jul 10, 2014 18:29:54 GMT -6
If you are using the Fade function in PT then it has a check box for adding dither. Right! Totally forgot about that. Thanks! |
|
|
|
Post by littlesicily on Jul 10, 2014 18:37:30 GMT -6
Hmmm… PT10.3.8… no "dither" option in fade dialog.  |
|
|
|
Post by mulmany on Jul 10, 2014 18:42:27 GMT -6
Hmmm... I am still on 9. What if you do power fade? Is there an option in setup?
|
|
|
|
Post by popmann on Jul 10, 2014 19:36:31 GMT -6
Reason 900 to not be using protools...I enable 32bit format in the project when it's time to mix. Everything stays 32bit float.
I seriously don't think it matters a lot when you're talking about a fade on an individual file, though. I'm not sure it matters "a lot" anywhere...on the final export, sure....why not...and the built in Apogee DOES have a 24bit setting...in fact, I don't think it HAS a 20 bit? I'm not sure-I've certainly never used it. What would that be for? Old 20bit DACs? So does IDR (have 24bit)--it's the default, no? At least on the plugs with IDR--the default is "on" and "24bit" on the mastering bundle stuff.
I really wouldn't overthink it. That would be missing my point--that I'm not sure how important it IS in practice, but there's simply no negative to doing so. If Waves defaults to IDR being on and 24bit--leave it. When you export a mix, insert dither as the last plug on the master set to whatever you're rendering to...
|
|
|
|
Post by popmann on Jul 10, 2014 19:38:38 GMT -6
But, for Cubendo users, that's a key thing--you can change the project bit depth on the fly. So, you TRACK with it set to 24bit...then flip it to 32bit when you go to do editing/mixing. That way, it's capturing the 24bit from the ADC and not wasting drive space...and then you flip it to 32bit (which doesn't convert anything already there) and do whatever rendering and editing you need to do--edits/renders are stored as 32bit float files.
|
|
|
|
Post by lolo on Jul 10, 2014 20:18:23 GMT -6
My head is spinning a bit here. So yeah after any processing dither a induvidual track or 2 buss when you bounce/print.
What if you dont print individual tracks and there is plugings or hardware inserts on them, do you still use a dither plug on the last insert? Sorry if this is a duuuumb question
Obviously still gonna dither on master buss before printing the Mix
|
|
|
|
Post by littlesicily on Jul 10, 2014 21:29:44 GMT -6
Hmmm... I am still on 9. What if you do power fade? Is there an option in setup? I looked in the prefs and didn't see any options for dither. |
|
|
|
Post by Bob Olhsson on Jul 10, 2014 22:22:25 GMT -6
I just set the session to 32 float when audio-suiting.
|
|
