|
Post by indiehouse on Jun 9, 2021 8:51:49 GMT -6
Speaking of dither, what’s best practices for using dither? Where and when? The rule always was any time there is a conversion. So before HW inserts? And before any bit depth/sample rate reduction bounces?
|
|
ericn
Temp
Balance Engineer
Posts: 14,936
|
Post by ericn on Jun 9, 2021 8:55:27 GMT -6
The rule always was any time there is a conversion. So before HW inserts? And before any bit depth/sample rate reduction bounces? That was always considered the best way to go, of course it comes down to how practical in the real world.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 8:55:59 GMT -6
The rule always was any time there is a conversion. So before HW inserts? And before any bit depth/sample rate reduction bounces? Yeah but you can’t dither float to float rounding.
|
|
|
Post by indiehouse on Jun 9, 2021 9:40:37 GMT -6
So before HW inserts? And before any bit depth/sample rate reduction bounces? Yeah but you can’t dither float to float rounding. I have no idea what that means.
Sometimes I feel like there's two types of audio "engineers". There's guys that talk in strictly in musical terms and guys that can talk in computer terms. Do I need to know how to do both? Do I need to know that I can't dither float to float rounding, or what the hell that even means?
|
|
|
Post by ragan on Jun 9, 2021 9:49:50 GMT -6
Yeah but you can’t dither float to float rounding. I have no idea what that means.
Sometimes I feel like there's two types of audio "engineers". There's guys that talk in strictly in musical terms and guys that can talk in computer terms. Do I need to know how to do both? Do I need to know that I can't dither float to float rounding, or what the hell that even means?
You absolutely don't need to know what that means or worry about it. This stuff is comically far-removed from the practical world of making records. I mean, I like it and I've chosen to study it, but it's not anything to spend any time on from the perspective of just recording and mixing music, in my view (unless you like it...). Have you ever read Zen and the Art of Motorcycle Maintenance? If not, the main character wants a deep harmony with every component of his bike, understanding exactly how each fits into the cohesive whole and creates the high-level experience of Riding A Motorcycle. His buddy can't stand the technical details. He's offended by them. For him, getting mired in that stuff greatly takes away from the whole ethos of riding. Anyway, they're both (of course) totally valid ways of approaching it, but unless you're going to design your own motorcycle (or DSP), you can happily cruise down the road without worrying about any of the gory details.
|
|
kcatthedog
Temp
Super Helpful Dude
Posts: 15,088
Member is Online
|
Post by kcatthedog on Jun 9, 2021 14:27:20 GMT -6
Way too esoteric for little ol me! Although, I liked the book !
|
|
|
Post by earlevel on Jun 9, 2021 15:06:12 GMT -6
Speaking of dither, what’s best practices for using dither? Where and when? Normally, the only time you need to be concerned is when you create a 16-bit version, for CD. Dither that truncation. No one ever ruined something by using good old TPDF dither, and at 16-bits and greater, I personally don't see the need for noise-shaped dither, but some like the theoretical aspects so have at it if you want. Unless you're doing something weird, with an unusual sample size crammed into a ROM for embedded, that's typically all that ever happens that needs dither. Of course, if you can always dither to 24-bit (say, from 32-bit float), it's just that no one will ever know or care, because all the error would be in the 24th bit and the electronics can't reproduce it even if your ears could hear it. To top it off, there's pretty much no way the recorded audio doesn't have a (relatively) huge amount of random noise, including Gaussian, that results in self-dither. Just saying, there are several reasons why 24-bit dither is completely unnecessary, each of which is ample many time over to ensure you can't tell the difference. Still, it's not hard (often a checkbox on export), so dither to 24-bit if you want (TPDF—noise shaping is full on lunacy at 24 bits, but since no one will hear that either, feel free to do that too if you must). So, in general, dither all reductions to int/fixed formats (mainly to 16-bit, and 24-bit doesn't matter either way, won't hurt, even though it's basically fixing a problem that doesn't exist in that case). People get mad when I don't just say "dither all reductions"—it's easy enough, and never "wrong". But there is a reason I make clear that it doesn't matter for 24-bit. Some people feel they must dither sends to external hardware (outboard reverb, etc.), or their music will be damaged. This is nuts, thermal noise alone will ensure that it makes no difference. Do it if you want, but I've seen people doing way to much fretting over whether they are dithering their sends correctly ("be sure there is no fader after the dither plugin!" etc.), for something impossible to hear, instead of making music. Makes me long for the days of analog tape where people just used it. And some people will say I can't know that—can't know if somehow our brains are so acutely attuned to that kind of distortion that they can pic it out of nothing, even when the energy isn't enough to deflect our ear drums (uh, and why would our brain even evolve with that peculiar speciality that doesn't exist in nature?). But this can be proved entirely objectively: Null the source material with the truncated material. You won't hear anything. Cheat and give 60 dB of digital gain or so to ensure your eardrum is engaged and can easily hear the difference. Now, A/B that with the null of the source and 24-bit dither version. Both will almost certainly sound like low level white noise that you can't tell apart. Yes, it's possible to compute a signal for which the truncation sounds bad, but without the cheat of digital gain to run it up into the higher bits of the DAC, it's be impossible for anyone to hear it.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 15:45:36 GMT -6
Using ITB attenuation for monitor level control is a method loaded with negative fidelity consequences. First, the obvious digital resolution reduction before reconstruction at the DAC. A common level attenuation for normal monitoring levels is somewhere is between -18dB to -26dB. This would reduce a 0dBFS peak to -26dBFP. But the average level of a mix lays down around -16 which, when attenuated, results in -40dBFS. That’’s only about 18bit resolution which will not reconstruct as faithfully when reconstructed at the DAC. We want faithfullness so that we know what’s in our record. Second, if dither is not being applied properly (which is most of the time) we are awarded two insults. First is truncation. The DAW may be 64bit but the DAC is only 24, so foul there. Next we inherit the cancer that is quantization distortion. This is a pesky distortion which modulates with the signal. Even when minuscule, the ear can pick this out of the noise floor and fake a mixer’s perceptions of their record. This leading us to make erroneous decisions. Third, without the benefit of an OTB level control, there is no stop guard for full scale casualties. When a computer has a “moment of confusion” and decides to pass full scale signal, there is no analog attenuation saving the amp from launching the speaker cones and, consequently, your ear drum. Safety first my sisters and brothers! Empirically, in my many blind listening tests over the decades, no ITB or built in DAC attenuation has been perceived as more resolute than a straight forward, external, high-quality, passive stepped attenuator switch. Yeah using the master fader as volume control is asking for disaster. Some daws don’t have inserts after it so you can’t dither the 64-bit float to 24-bit truncation and prevent distortion. It’s so easy mixing itb to make the daw hang if you forget to freeze with the best modern plugs. Sometimes the interface really does sound better at -1dbfs (MOTU) to even -3 to -6 (Steinberg). It just depends on what causes the least distortion.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 15:52:25 GMT -6
Speaking of dither, what’s best practices for using dither? Where and when? Normally, the only time you need to be concerned is when you create a 16-bit version, for CD. Dither that truncation. No one ever ruined something by using good old TPDF dither, and at 16-bits and greater, I personally don't see the need for noise-shaped dither, but some like the theoretical aspects so have at it if you want. Unless you're doing something weird, with an unusual sample size crammed into a ROM for embedded, that's typically all that ever happens that needs dither. Of course, if you can always dither to 24-bit (say, from 32-bit float), it's just that no one will ever know or care, because all the error would be in the 24th bit and the electronics can't reproduce it even if your ears could hear it. To top it off, there's pretty much no way the recorded audio doesn't have a (relatively) huge amount of random noise, including Gaussian, that results in self-dither. Just saying, there are several reasons why 24-bit dither is completely unnecessary, each of which is ample many time over to ensure you can't tell the difference. Still, it's not hard (often a checkbox on export), so dither to 24-bit if you want (TPDF—noise shaping is full on lunacy at 24 bits, but since no one will hear that either, feel free to do that too if you must). So, in general, dither all reductions to int/fixed formats (mainly to 16-bit, and 24-bit doesn't matter either way, won't hurt, even though it's basically fixing a problem that doesn't exist in that case). People get mad when I don't just say "dither all reductions"—it's easy enough, and never "wrong". But there is a reason I make clear that it doesn't matter for 24-bit. Some people feel they must dither sends to external hardware (outboard reverb, etc.), or their music will be damaged. This is nuts, thermal noise alone will ensure that it makes no difference. Do it if you want, but I've seen people doing way to much fretting over whether they are dithering their sends correctly ("be sure there is no fader after the dither plugin!" etc.), for something impossible to hear, instead of making music. Makes me long for the days of analog tape where people just used it. And some people will say I can't know that—can't know if somehow our brains are so acutely attuned to that kind of distortion that they can pic it out of nothing, even when the energy isn't enough to deflect our ear drums (uh, and why would our brain even evolve with that peculiar speciality that doesn't exist in nature?). But this can be proved entirely objectively: Null the source material with the truncated material. You won't hear anything. Cheat and give 60 dB of digital gain or so to ensure your eardrum is engaged and can easily hear the difference. Now, A/B that with the null of the source and 24-bit dither version. Both will almost certainly sound like low level white noise that you can't tell apart. Yes, it's possible to compute a signal for which the truncation sounds bad, but without the cheat of digital gain to run it up into the higher bits of the DAC, it's be impossible for anyone to hear it. The null test is pointless. Truncation distortion causes additional unwanted harmonic distortion that may cause you to process more. You’re not hearing the truncation distortion itself but the results of its effect on the partials in the signal. To presume we can’t hear it is to cut yourself off from valid and tested human perceptions of multiplicity. I have heard numerous examples of notes that disappear into the background when an additional truncation event is present. The string might not have been hit. You can’t null things like that. Then you’re like the guys nulling eqs 10 years ago on Gearslutz without considering the impulse responses and phase. Anything not exactly the same is different. That’s all a null test proves. And dithered vs truncated does not null. EQs with the same eq filters and precision but different oversampling filters will not null.
|
|
|
Post by earlevel on Jun 9, 2021 16:28:23 GMT -6
The null test is pointless. Truncation distortion causes additional unwanted harmonic distortion that may cause you to process more. You’re not hearing the truncation distortion itself but the results of its effect on the partials in the signal. To presume we can’t hear it is to cut yourself off from valid and tested human perceptions of multiplicity. I have heard numerous examples of notes that disappear into the background when an additional truncation event is present. The string might not have been hit. It's hard to understand what you're claiming here, in an objective sense. We can "null things like that". TPDF dithered output is sample-perfectly in phase with the input. Sorry, I don't know you level of expertise, I don't want to sound condescending by explaining how it works, but matching phase is not an issue in this case—the samples line up perfectly, you can do this test easily in a DAW in real time, or in an audio editor, with no adjustment. From there, the math is linear: A_sig - B_trunc = C_null, then B_trunc = A_sig - C_null. In other words, the truncated signal is precisely the original signal summed with the inverted null. And this simple math stops working once ears are involved. For 24-bit it's still true, it's just that you will never hear it without artificially fixing the game (by moving the signal up to higher bits digitally. Analog gain won't help because you'd be amplifying the noise of the electronics).
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 18:39:17 GMT -6
At the end or the day it’s either low level noise or distortion. You can pretend the distortion doesn’t exist but it’s there. I don’t have time to argue any more over this tonight. I have real work to do on this record. Adieu for now.
|
|
|
Post by indiehouse on Jun 9, 2021 19:14:18 GMT -6
At the end or the day it’s either low level noise or distortion. You can pretend the distortion doesn’t exist but it’s there. I don’t have time to argue any more over this tonight. I have real work to do on this record. Adieu for now. What are you using as a monitor controller?
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 19:33:35 GMT -6
At the end or the day it’s either low level noise or distortion. You can pretend the distortion doesn’t exist but it’s there. I don’t have time to argue any more over this tonight. I have real work to do on this record. Adieu for now. What are you using as a monitor controller? I just use the MOTU webpage mixer. It sounds better at -1 anyway so whatever digital distortion it may be doing is already there (-1 to passive beats 0 to passive) if you do an external volume control and the Alps and JBL (Chineseium I guess) pots I have degraded the sound further. -1 The Drawmer was imposing. Dangerous Source was very imposing and lacked monitor control features other than a headphone amp and a Bournes pot. I’m too cheap for a Coleman right now but that’s what I want to try when I get an interface or converter that sounds best at 0 dbfs.
|
|
|
Post by earlevel on Jun 9, 2021 19:50:36 GMT -6
At the end or the day it’s either low level noise or distortion. You can pretend the distortion doesn’t exist but it’s there. I don’t have time to argue any more over this tonight. I have real work to do on this record. Adieu for now. We don't have to argue. I'm fine with however anyone likes to work. Of course, I'm going to have a reply if someone (not you, just giving an example) says something like "I can hear the 24th bit". Two legend mastering engineers whom I have great respect for have told me this—still respect them, but they can't. Another grammy-winning friend wanted to send me an audio file by FedEx, because sending it over the internet "changes the bits, it sounds different". OK, he's still a great engineer. I'm not trying to change anyone's mind, just addressing the technical facts, let people decide. I'm only a little hard-line on this topic because one of those legends has told people if they don't dither their sends, their work will be ruined. And since it's slightly more than trivial (people new at it are inevitably paranoid that they might be getting a digital gain change after the dither, and fear they are subtly making things even worse). But if anyone reading wonders if the 24-bit dither/distortion level can or can't actually be heard, on their own equipment by their own ears, I can share a 2-bit digital signal that is very easy to identify, if you can hear it at all. LMK
|
|
|
Post by indiehouse on Jun 9, 2021 20:00:18 GMT -6
What are you using as a monitor controller? I just use the MOTU webpage mixer. It sounds better at -1 anyway so whatever digital distortion it may be doing is already there (-1 to passive beats 0 to passive) if you do an external volume control and the Alps and JBL (Chineseium I guess) pots I have degraded the sound further. -1 The Drawmer was imposing. Dangerous Source was very imposing and lacked monitor control features other than a headphone amp and a Bournes pot. I’m too cheap for a Coleman right now but that’s what I want to try when I get an interface or converter that sounds best at 0 dbfs. Man, that’s risky. I’d be liable to accidentally blow my speaker drivers. Or my ears.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 9, 2021 20:05:35 GMT -6
I just use the MOTU webpage mixer. It sounds better at -1 anyway so whatever digital distortion it may be doing is already there (-1 to passive beats 0 to passive) if you do an external volume control and the Alps and JBL (Chineseium I guess) pots I have degraded the sound further. -1 The Drawmer was imposing. Dangerous Source was very imposing and lacked monitor control features other than a headphone amp and a Bournes pot. I’m too cheap for a Coleman right now but that’s what I want to try when I get an interface or converter that sounds best at 0 dbfs. Man, that’s risky. I’d be liable to accidentally blow my speaker drivers. Or my ears. You’re already risking blowing stuff up with every crash and pop. This is why sturdy drivers are important. When I tried using the Dangerous Source over Usb, it crashed so much, I was afraid for my speakers. Buffer issues.
|
|
|
Post by indiehouse on Jun 9, 2021 20:14:46 GMT -6
Man, that’s risky. I’d be liable to accidentally blow my speaker drivers. Or my ears. You’re already risking blowing stuff up with every crash and pop. This is why sturdy drivers are important. When I tried using the Dangerous Source over Usb, it crashed so much, I was afraid for my speakers. Buffer issues. I guess I’m lucky in that department. Everything has been super stable for me for a while. I’m on a Mac.
|
|
|
Post by jmoose on Jun 9, 2021 22:15:55 GMT -6
I have no idea what that means.
Sometimes I feel like there's two types of audio "engineers". There's guys that talk in strictly in musical terms and guys that can talk in computer terms. Do I need to know how to do both? Do I need to know that I can't dither float to float rounding, or what the hell that even means?
You absolutely don't need to know what that means or worry about it. This stuff is comically far-removed from the practical world of making records. I mean, I like it and I've chosen to study it, but it's not anything to spend any time on from the perspective of just recording and mixing music, in my view (unless you like it...). It's valuable to know at least some theory, the nuts & bolts of how and why things work the way they do. But ultimately right... the propellerhead stuff is almost completely removed from the process of actually making records and can absolutely get in the way of that. In more practical terms... indiehouse This is kinda what I was getting at in one of your converter threads. That is, how do you know what your "actually" listening to and evaluating? Are you hearing the converters? Or are you also hearing monitor management? It can get convoluted for sure. Fwiw im using the volume control built into the SSL desk. If & when I've got a remote gig I'll use the volume control on the interface and not sweat it. Sounds fine. Flip side, splitting hairs. While waiting for the SSL to arrive last year I had to mix a couple projects with my shop torn apart and used monitors plugged right into the interface. People have deadlines I have bills. Can't stop working. When the desk finally arrived, I plugged outs 1&2 into the 2 track return... Same monitors and hail Satan what a difference in sound. Like oh. Yeah. That's what I was missing! Way more depth and detail in the soundstage.
|
|
|
Post by Ward on Jun 10, 2021 7:03:46 GMT -6
Might I ask what is everyone's preferred recording level before attenuation? I tend to stick to averages around from -21 dBfs (digital -3) to -18dBfs (digital 0). Peaks get as high as -3dBfs on some things but then I adjust downwards because you can't unclip a transient or program material.
I have friends who fry the converters with way too hot signals.
So I'm curious what the general consensus is here . . . in 2021 going forward.
|
|
|
Post by Guitar on Jun 10, 2021 7:31:26 GMT -6
I haven't caught up on the thread yet but I'm usually, on tracks, somewhere between -18 and -3 ish like you said.
When things get too hot sometimes I have to stop and cool everything down, then proceed. It's like everyone is shouting sometimes.
The real hard one for me is mastering levels, but that's a whole other topic. I'm starting to ease up on loudness a bit, lately.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 10, 2021 7:44:23 GMT -6
Might I ask what is everyone's preferred recording level before attenuation? I tend to stick to averages around from -21 dBfs (digital -3) to -18dBfs (digital 0). Peaks get as high as -3dBfs on some things but then I adjust downwards because you can't unclip a transient or program material. I have friends who fry the converters with way too hot signals. So I'm curious what the general consensus is here . . . in 2021 going forward. I just make sure not to clip anything and always trim before mixing unless I am running everything into shit like U-He Satin with the soft clipper on. I certainly abuse that one.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 10, 2021 10:13:20 GMT -6
I have no idea what that means.
Sometimes I feel like there's two types of audio "engineers". There's guys that talk in strictly in musical terms and guys that can talk in computer terms. Do I need to know how to do both? Do I need to know that I can't dither float to float rounding, or what the hell that even means?
You absolutely don't need to know what that means or worry about it. This stuff is comically far-removed from the practical world of making records. I mean, I like it and I've chosen to study it, but it's not anything to spend any time on from the perspective of just recording and mixing music, in my view (unless you like it...). Have you ever read Zen and the Art of Motorcycle Maintenance? If not, the main character wants a deep harmony with every component of his bike, understanding exactly how each fits into the cohesive whole and creates the high-level experience of Riding A Motorcycle. His buddy can't stand the technical details. He's offended by them. For him, getting mired in that stuff greatly takes away from the whole ethos of riding. Anyway, they're both (of course) totally valid ways of approaching it, but unless you're going to design your own motorcycle (or DSP), you can happily cruise down the road without worrying about any of the gory details. Every slight/minute improvement has built up into better sound for me. I have to internalize it into routine practice for real world use. I need teach myself to use the better tools. These might not be the best measuring ones (monitors and processors) but the ones that measure good enough and push you or let you do what needs to be done, eg think Blockfish vs Renaissance Compressor back in the day. And this is outside of the real world of making records. You can’t measurebate and experiment on the clock.
|
|
|
Post by Guitar on Jun 10, 2021 10:16:33 GMT -6
That's how I approach it too. "Theory" in the off time, for fun and education, calibrate the tools, build some things. When working, no thought of it whatsoever, just make music. Very much a separation of the two.
|
|
ericn
Temp
Balance Engineer
Posts: 14,936
|
Post by ericn on Jun 10, 2021 10:29:13 GMT -6
That's how I approach it too. "Theory" in the off time, for fun and education, calibrate the tools, build some things. When working, no thought of it whatsoever, just make music. Very much a separation of the two. The first thing you learn is that in the land of theory every other variable is a constant. In the real world nothing is a constant or equal!
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Jun 10, 2021 16:53:05 GMT -6
That's how I approach it too. "Theory" in the off time, for fun and education, calibrate the tools, build some things. When working, no thought of it whatsoever, just make music. Very much a separation of the two. The first thing you learn is that in the land of theory every other variable is a constant. In the real world nothing is a constant or equal! Certainly not anything non-linear! Every dynamics processor has nutty peculiarities. Oxford Dynamics basically changes the drum performance and claims to be clean! It makes real drums sound like hard hitting samples if used at any faster attack or release. A DBX 160 “accidentally” turns into an expander before it clamps down! The great majority of fast attack or release digital compressors are pretty much random dysfunctional shit boxes it’s up to the user to make sound cool because they don’t work on half the settings or any of them!
|
|