|
|
Post by ragan on Aug 26, 2017 20:20:40 GMT -6
Hey so I just did another round of blind AB'ing those same files I posted earlier and I picked the dithered version every time. I'm not putting them into an ABX app or anything, I just set it up so I can swap back and forth with a keystroke and close my eyes and start going back and forth and when I come up with a preference, open my eyes and see which it is. Almost felt like I was making it up, but if so, I just 'made up' a preference for the dithered version about 8x in a row. Just a tiny bit of added clarity. A minuscule amount, but I kept picking the dithered version. The other night, I straight up couldn't come up with a preference at all. So definitely gonna try it some more with more channels and on other sources, just thought I'd share. Edit: and then now I just did a few where I picked the undithered version and again felt like I was just kinda guessing. Jury's out for me but I'm definitely gonna try a bunch more comparing. Edit II: doing some drum OH, now I'm back to random guessing, picking the dithered sometimes, undithered sometimes, can't hear any difference right now I think your ear is becoming educated to this particular issue. And AB testing tends to bias the ear towards confusion, as I noted previously. Yeah, could be. |
|
|
|
Post by mulmany on Aug 26, 2017 21:36:51 GMT -6
When I tested myself I remember "feeling" more comfortable with the dithered HW insert. So I have taken the Bob's advice and incorporated it into my work flow. I also watched Johnston's AES talk on dither and decided to go with the mathmatacly correct version. It does not hurt anything to do it, and this way I don't have to ever wonder if I should do it.
|
|
|
|
Post by donr on Aug 27, 2017 1:15:13 GMT -6
Further correspondence with Dave Roberts at MOTU. He says coming out of DP to an interface at 24 bits fixed from 32 bit float involves no truncation and no advantage to adding dither.
I wonder if what people prefer using dither is the noise. Like a painting on black velvet rather than canvas. Why do so many modeled plug-ins add noise? The noise floor of vinyl records is like -55dB, isn't it?
|
|
|
|
Post by Bob Olhsson on Aug 27, 2017 15:36:26 GMT -6
That's a very common misconception among developers. It's true 32 float is only, if I remember right, 23 bits of data however it's calculated with every sample at full level. When you convert this back to fixed point in order to drive a DAC, there will still be truncation distortion unless you dither.
|
|
|
|
Post by ragan on Aug 27, 2017 16:03:32 GMT -6
Further correspondence with Dave Roberts at MOTU. He says coming out of DP to an interface at 24 bits fixed from 32 bit float involves no truncation and no advantage to adding dither. I wonder if what people prefer using dither is the noise. Like a painting on black velvet rather than canvas. Why do so many modeled plug-ins add noise? The noise floor of vinyl records is like -55dB, isn't it? Think Dave would be interested in chiming in here? |
|
|
|
Post by EmRR on Aug 27, 2017 17:01:17 GMT -6
Further correspondence with Dave Roberts at MOTU. He says coming out of DP to an interface at 24 bits fixed from 32 bit float involves no truncation and no advantage to adding dither. I wonder if what people prefer using dither is the noise. Like a painting on black velvet rather than canvas. Why do so many modeled plug-ins add noise? The noise floor of vinyl records is like -55dB, isn't it? So far Dave hasn't chimed in on my multiple musing at MOTUnation about this. I agree he sounds wrong on this one, but read on for variables: Particularly that the AVB interfaces receive 32 bit words, and Audio MIDI Setup on a Mac describes the 16A as 32 bit-float. The scant data on the D/A converter chip mentions the output trims are digital and use the DAC's internal trim, and the filter architecture is also 32 bit. Is there dither here? Am I double dithering by using a dither plug in post-fader in DP? BTW, I run the outputs -4 so they are closer to unity in a loop situation, so again, am I truncating again within the DAC or is dither applied appropriately? I've been running the native MOTU dither plug on all outputs, post-fader, for awhile. I can't fault it, impression is a bit less glare/hardness as sessions progress, but I haven't shot it out. Bob is absolutely right about the question of subsequent processing....like the impact of mastering on your truncated files, a result you can't predict until too late. Dither certainly won't hurt! Math is fact, I tend to trust the Bell Labs statement (smartest guys on the planet), and I don't see any point in making it a religious issue. Just do it since the manufacturers/designers screwed up the implementation along the way. Maybe they'll steer the ship back right, but entropy doesn't tend to go that way. |
|
|
|
Post by ragan on Aug 27, 2017 21:13:41 GMT -6
There certainly is an interesting point to be made there re: what's the downside?
I print hardware a few channels at a time. The tiny extra 'work' of throwing ProL or some other thing on before the HW insert is meaningless to me. I guess I'll just do it for now. I'd love a definitive answer on the math side though whether truncation is taking place or not.
I'm about to take like 2 years of straight math so hopefully one day I'll have a better understanding of these things.
|
|
|
|
Post by EmRR on Aug 27, 2017 21:35:25 GMT -6
I'd love a definitive answer on the math side though whether truncation is taking place or not. The definitive answer is that if any processing ( fader change counts) has occurred, the word length has extended and is then truncated UNLESS you can confirm your DAW automatically dithers outputs at 24 bit, or your converter (I know of no evidence proving this to occur so far). |
|
|
|
Post by jeromemason on Aug 27, 2017 21:52:31 GMT -6
I've been reading this trying to get a bearing on what this is all about and how I would use it.
So, if I am mixing a project that is 44/24 and I have 4 busses that are sent out of my Motu 16a and ran through my analog buss chain, does this mean I need to be dithering every track that's routed to it's respective analog buss, and then on my PRINT return track I should dither this as well?
|
|
|
|
Post by lcr on Aug 28, 2017 8:51:00 GMT -6
^^ depends on who you ask. I suggest you do a comparison with and without dither and see if you notice a difference.
|
|
|
|
Post by swurveman on Aug 28, 2017 9:17:46 GMT -6
To add to the mystery here is Paul Erlandson of Lynx A/D converter Studio Technology Support saying, "DAWs that operate at 32-bit internally still deliver 24-bit words to the Audio Device. No dither is necessary." Dither |
|
|
|
Post by ragan on Aug 28, 2017 9:41:21 GMT -6
Does it not seem like the MOTU guy and the Lynx guy would, you know, have a relatively good handle on this?
I'm more curious than ever.
|
|
|
|
Post by lcr on Aug 28, 2017 10:53:23 GMT -6
If anything solid is concluded, I really hope no ones feelings get hurt. Im looking forward to having the time to do my own test.
|
|
|
|
Post by ragan on Aug 28, 2017 11:08:46 GMT -6
I think we should all invest a helluva lot less of our fee fees (read: IDENTITY) in things like this.
|
|
|
|
Post by jeromemason on Aug 28, 2017 12:16:46 GMT -6
^^ depends on who you ask. I suggest you do a comparison with and without dither and see if you notice a difference. Eh, man I don't have time to do that, I've got to start mixing an EP this afternoon and if I need to be doing this I'll do it, but if all I'm doing is adding noise that's the last thing I want to do. I guess I could just do the mixes and print a with and without on each song and let the producer decide, that'd be interesting, take me out of the equation. |
|
|
|
Post by lcr on Aug 28, 2017 12:22:45 GMT -6
^^ do this. Im sure Im not the only one following this discussion that would like to hear the verdict. Not actual files, just if you or anybody noticed a difference.
|
|
|
|
Post by EmRR on Aug 28, 2017 12:23:22 GMT -6
If anything solid is concluded, I really hope no ones feelings get hurt. Im looking forward to having the time to do my own test. Well, Bell Labs is as solid as it gets. That doesn't mean people won't do the wrong thing, and it won't matter. No one lives or dies by audio. |
|
|
|
Post by EmRR on Aug 28, 2017 12:26:28 GMT -6
^^ depends on who you ask. I suggest you do a comparison with and without dither and see if you notice a difference. but if all I'm doing is adding noise that's the last thing I want to do. implying that all one is doing is adding noise shows how confused people are by this. There's plenty of factual info in this thread, and many others. If you want to weight the responses of armchair theorists equally, then you might think it's open to debate. But, again, no one lives or dies by it, so it really doesn't matter. It's damn well easy enough to add everywhere and never think about again. |
|
|
|
Post by EmRR on Aug 28, 2017 12:27:38 GMT -6
Im sure Im not the only one following this discussion that would like to hear the verdict. There is a verdict. It's being re-debated for some reason. It's not necessarily about what you hear NOW. |
|
|
|
Post by ragan on Aug 28, 2017 12:29:02 GMT -6
If anything solid is concluded, I really hope no ones feelings get hurt. Im looking forward to having the time to do my own test. Well, Bell Labs is as solid as it gets. That doesn't mean people won't do the wrong thing, and it won't matter. No one lives or dies by audio. Right but isn't Bell Labs or whoever's statement based on the idea that there is in fact truncation? MOTU and Lynx guys claim there isn't actually truncation and it's still 24 bit words at the device level. I certainly don't know that anser re: 32 bit float/24 bit ADC but doesn't it all hinge on that? Then the question of audibility and noise floor and all that come into play, but doesn't the theoretical math question hinge on the truncation/word length question? |
|
|
|
Post by johneppstein on Aug 28, 2017 12:38:27 GMT -6
Further correspondence with Dave Roberts at MOTU. He says coming out of DP to an interface at 24 bits fixed from 32 bit float involves no truncation and no advantage to adding dither. I wonder if what people prefer using dither is the noise. Like a painting on black velvet rather than canvas. Why do so many modeled plug-ins add noise? The noise floor of vinyl records is like -55dB, isn't it? That's kinda shocking. It reveals something about the depth of ignorance that exists even at respected manufacturers. No truncation from 32 bit float to 24 fixed? Can this guy even do basic arithmetic? Linkedin says he's a "customer relations manager/product specialist". NOT somebody I would quote as a scientific authority, especially when what he says contradicts a guy like J_J. |
|
|
|
Post by johneppstein on Aug 28, 2017 12:47:04 GMT -6
I just invited J_J to drop in on this thread, hopefully he's not too busy.
|
|
|
|
Post by jeromemason on Aug 28, 2017 13:01:15 GMT -6
^^ do this. Im sure Im not the only one following this discussion that would like to hear the verdict. Not actual files, just if you or anybody noticed a difference. Yeah I will. I can't post any type of files obviously but I'll update the thread on his choices. |
|
|
|
Post by EmRR on Aug 28, 2017 13:19:13 GMT -6
there is in fact truncation This yeah, you didn't say that exactly, but I'm highlighting that part. Think about it. Why on earth would they give us a 24 bit dither option at all, if it didn't matter? 32 float is indisputably more data than 24. It is truncation, with all it's attendant distortion. Dither helps this. |
|
|
|
Post by ragan on Aug 28, 2017 13:25:55 GMT -6
there is in fact truncation This yeah, you didn't say that exactly, but I'm highlighting that part. Think about it. Why on earth would they give us a 24 bit dither option at all, if it didn't matter? 32 float is indisputably more data than 24. It is truncation, with all it's attendant distortion. Dither helps this. Well again, I straight up have no idea. I know you know what you're talking about, but I'm also curious about Lynx guy's seeming distinction of word length being "delivered to the device". And again, I have no real 'view' here because it's way over my head. Just curious. If there truly is no question, then yeah, seems the obvious 'best practice' is to dither. Whether or not the affect is audible, that would be the mathematically correct thing to do. I should say too, I'm very familiar with the sort of 'pretend' debates. My wife is a rhetoric of science professor and she's made a career out of studying such things. I realize the things I'm wondering about may be solidly and inarguably established. I'm just on a curiosity mission. My mind goes "well, Doug knows what he's talking about, he says there is truncation, Lynx guy ostensibly knows what he's talking about, he says there isn't." My mind can't help but want that to be resolved. That's all. |
|
