To be honest, dithering or not at 24 bit doesn't matter that much. When I tested this ages ago I heard a noticeable improvement with dithering. I don't understand why people don't dither even if they can't hear a difference. It's technically and mathematically the right thing to do and good practise. Not dithering has zero benefit. You said dithering is already occuring between file handling protocols. Basically, in most daws if you export a 16 or 24 bit file you can specify to use dither or not. Dither in floating point is incorporated into some of Sonnox plugins I think. But I don't quite understand this, as I didn't think you could dither to floating point. Anyway, to cut a long story short, you usually have to explicitly choose dither when converting bit depth. In old Pro Tools TDM, which worked in fixed point, there was a choice to use a dithered or undithered mixer. I don't know why anyone would choose the undithered one though. Anyone know if the undithered one had any benefit?

From what I've picked up from this thread, it seems dithering isn't necessarily just for bit reduction, ideally one should also dither when rendering to the same fixed-point bit depth too, ie. 24bit -> 24bit. I also monitor through the dither I will be using for the final mixdown. The only time I don't dither is when rendering audio to floating point.

These days I just make sure the DAW's own dither is turned off for rendering and playback, then I work and bounce at 32bit float throughout, with dither as the last insert on my monitoring bus (rendered audio doesn't go through the monitoring bus.) Once the mix/master is finished, I bounce it down to 24bit fixed point through the dither as the last insert. If I'm syncing the master to video, I will keep it 32bit float and import that to Final Cut, then add the dither there before rendering the finished product, usually to 24bit fixed point.

I will tread carefully here as I don't wish to mislead, but through some other discussions elsewhere I'm now aware of what the Sonnox does, which is not the same thing as what I do: and I think I can outline who does what and why. Far as I know it is only me and Paul Frindle, and we've got significantly different approaches held to the death so that's why I tread carefully.

I believe what the Sonnox stuff is doing is this: in floating point, there is half the waveform (before clipping of the output buss) that's all at one fixed point level. The mantissa for all samples less than 0dB but louder than -6dB (point etc etc etc: yes it's not exactly 6, it's exactly 1/2) is something, I think it's around 25 bit? The point being half your waveform swing, including the outer edges of the waveform of bass notes where you'll want linearity, is a specific fixed point amount, as the mantissa is a predictable factor.

So what Sonnox is doing, and it's Paul behind it and he had this long before I got interested in the subject, is setting ONE dither level, and dithering correctly for the outer half of the waveform. If you went past 0dB it would be an inadequate level, if you have quieter sounds the tiny quantization is not being dithered as the dither signal is then too strong, but you know with the quieter sounds and excessive dither signal what happens is mostly the noise taking on an objectionable quality. It's still noise, and we're talking _25_bit_ dither noise here: not a problem.

And the reason that choice is made is (a) for literally half the waveform swing you're dithering the mantissa correctly for perfect results and (b) if you're not, it's not a problem and you're not fluctuating the amplitude of the dither. Your dither is unvarying in level, like TPDF dither should be. I think this is reasonable. I don't know anybody but Paul/Sonnox who've arrived at this.

I ran with a different approach, scaling the dither to match the mantissa. Works for me. The reason I like it is, if you compress heavily or combine lots of very faint sounds, any bit of audio at any level being transmogrified to any other level, whether it's super-faint or super-loud or any combination, every mantissa will always be linearized the same: yes, the noise of that will scale with the sound, but I'm more interested in providing the system with dithered audio regardless of how it might later be scaled. With mine if you blow through a section of processing with your audio hitting +47dB peaks and then being trimmed back later, it still works, and if you do a bunch of stuff 80 dB down and then bring it up, it's got dither noise appropriate to the source amplitude. Which is inaudible in any one stage, because it's at 25 bit at worst, and probably much more finely grained than that, and dithered: none of this will ever be heard by anybody.

And again, I've got test files where I've run a mix with long double variables and one with all the variables as floats, and you absolutely do hear a textural/spatial difference. The sum total of all the inaudible differences is ABXable, sufficiently obvious to stand up to rigorous testing. Any one of the inaudible differences, by itself, is not. No human could hear the difference on one instance of this, put 'em together and I think many people would be able to pass an ABX test with high confidence. Including some of the folks who are on the other side of the argument, they've got ears too y'know.

It sounds interesting, though I'm not convinced it would sound better to dither to float. Definitely fixed, but I'd need to be convinced on float.

Candyflip, you don't dither 24 bit to 24 bit. Only dither if it's ever expanded to float, which is any process including volume level change.

...or sample rate conversion. You'll end up with all new samples and sample values

Yes, I wasn't clear on that part sorry, I was meaning that when 24bit fixed audio is adjusted in any way (more often than not by plugins/DAW at float) and rendered to 24bit fixed, it is wise to use dither. So you are indeed correct, technically bit reduction is being done in between renders. Thanks!

"Your reality"...OK, at least it's reality.

I'm not "imposing" anything. I've said more times than I can count that people can do whatever they want. It's you who want my voice to be gone. I don't want yours to be gone, so I find that puzzling. I'm just giving the info. If it saves someone from anguishing over the realization that they didn't dither their sends in a project they just mixed, that would be great. you'd prefer they at least worry about it, if not redo it—or at least that's the impression I get.

No matter how much you distain science on a "music forum" (it's a gear forum, I haven't seen a discussion of music in this thread, I've seen discussion of digital audio and signal processing, and electronics gear), some other people who read this just might appreciate it. If you believed I'm wrong, you could just pull an electrical engineer of physicist in here to tell us why I'm wrong. Oh, wait, you don't like science guys...

OK, just understand that even on a cold day, there's a lot of heat in the air, causing molecules to bounce around—you will never encounter a day in your life without that, because you won't ever be at absolutely zero. Similarly, there is "bouncing" going on in any gear you've ever bought or ever will—find me anything with an snr greater than 130 dB if you think I'm wrong, no scientist needed. And there are quantum effects—5000 electrons is half 10000 electrons, but there is nothing that's half an electron, so once you get to tiny current flows, it's in discrete steps, and you can't accurately render continuous changes—shot noise. And since you can't go less that 1 without going to zero, it should be obvious that you can't just use more bits to encode smaller and smaller signal fluctuations.

All of nature has limits—you can't see microbes because your retina can't resolve them, and even if it could, the wavelengths of visible light can't resolve them. It's true no matter how badly you hate being told the world we live in has limits.

I dont want anyone to stress if they didnt dither in the past. I did many thousands of masters before I started using 24 bit dither on the feed to my monitor DAC. NBD.

Yes, the physical world has limits. Without limits, science wouldn't exist. This is why science fails to explain the limitless or the infinite, but we pretend it can because it would be too scary of world otherwise. It is much safer for the mind to obsess on easy problems (like dither) than it is on the hard problem.

I dont want you to leave or stop talking about the science of digital audio. I find it fascinating, and you obvious know your stuff. Where I draw the line, over and over again, is when you consume the nature of conscious perception into your materialist (small problem) worldview.

Why not just let people's perceptions be?

Dont belittle as the knowledge authority, dont judge as biased listening, dont condescend with reductionism, just take a deep breath and say "it is OK if someone else experiences that which my mind says is impossible to experience."

No threat need be taken from what I or anyone else experiences.

All reasonable, Robb, thanks. I think I am "letting peoples' perceptions be"—just giving some other info to consider. I don't think I belittle anyone's knowledge. I recognize it may sound a little like that at times, and doubly after someone has called it preposterous (not implying you did, but obviously I've had to defend my claims a few times, just like those who rely on listening have), but there is no other way to say there are hard limits without just saying it. But I really don't care what people do with the info, I'm just presenting it the bast I can. If someone else can be adamant about what they hear, I think it's fair that I can be adamant about what I know, measure, and perceive as well.

And again, if we were only talking about dithering the final product, I've said over and over—"do it—why not? There's practically no effort, it won't hurt anything, might as well feel comfortable about it." But there are cases, for some people, where it helps to know these things. I think you view it as me trying to change the minds of the whole board, whereas I always figure I'm talking to maybe one or five people over the course of how many years this thread will be viewed. Also, I know for sure there are people who read this and know what I say is true. But there is no benefit in saying so out loud.

So, it boils down to "I gotta be me", and "you gotta be you", as it should be, even though neither will benefit from, or move the needle to improve mankind's lot, on an advertising-fueled message board. We're just chatting.

And I truly appreciate you taking the time to make your thoughts clear.

We are very fortunate to be able to participate in a robust and respectful discussion, involving so many people from different backgrounds, locations and experiences.

...and all out of LOVE for the music!!!

While many scientific/mathematical facts have been offered in this thread, I believe it is important to remember that, at the end of the day, the scientific/mathematical theories/rules/descriptions are snippets of information by which we seek to understand complex real-world phenomena.

This is not to say that the theories are false or even 'lacking' in themselves, but rather, that we must take caution when claiming that the 'snippets' themselves can completely explain or deny the existence of real-world effects; which are usually incredibly complex and multi-faceted.

In this sense, from a scientific perspective, I believe it would be very worthwhile to investigate the mechanisms by which dithered audio can possibly affect human perception, especially at seemingly 'imperceptible' levels.

For example, perhaps 'level' (i.e. amplitude), viewed in isolation, is not the main driving factor in the 'perceived' result for dithered audio.

Personally, I find fascinating the degree of consistency in the descriptions of dithered vs non-dithered audio (at say, 24 bit), from those who claim to perceive (at times) a difference between the two. (Note: this is not to imply that every person does, can or should perceive these differences, or perceive them in the same way).

One possible conclusion is, of course, "it's all in their head/s", or eyes, or wallets, or 'forum hysteria'...

There could, however, be other conclusions.

That's basically the history of Science itself. Sometimes the solutions to such situations are remarkably simple, just not well understood at a given point in time.

And we can all still make great music, irrespective of those conclusions!

I think that it's just possible to hear below noise, that's probably why 24 bit dither can be audible, even though the noise floor might be higher. If you have scientific evidence for why this is impossible I'm all ears.

So I might be doing this test wrong, but what I tried was:

White noise generator at -120dB to be the noise floor that anything under we supposedly can't hear or something. Then a random song that's limited and peaking at 0dB, then turned down -138dB (the level of 24 bit dither). Now I can't hear any of this until I turn the master up 20dB. But not listening very loud. But if I turn it up some more, the point is I can hear the song under the white noise. So doesn't this mean 24 bit dither is still audible under -120dB of noise? I think the question of whether 24bit dither is audible is more to do with dynamic range of human hearing, which is about 120dB. So saying something is covered by noise just doesn't work for me. I do understand the argument that it's too quiet to hear. But I still think when sounds are so quiet like at -140, they might still be audible, but maybe more subliminally. Also, maybe hearing quiet sounds is easier when there are much louder sounds with it. Maybe I'm just waffling now, but I think the argument to be had about 24 bit dither is about how quiet humans can hear. Not anything about noise being louder and concealing it. And remember that 24 bit dither is often not the last process, so limiting after it is going to easily make it audible

I just did a test where I mixed some Gaussian noise (which is the most similar to my line input noise) with a 440Hz sine wave. The sine wave peak was 40dB lower than the noise peak and it was still audible! The RMS level of the noise was also significantly higher than the signal.
I don't know how well this simulates an actual DAC/ADC noise scenario, but it's interesting.

Right, of course. Summing up all the noise frequencies you get the total noise amplitude, which is significantly higher.

But, as I said, I don't think the total noise (SNR or dynamic range) is a very good indicator of what's audible, or of what's the lowest signal that can still be (partially) reproduced.
So, a -150dB peak signal from that device should still show up on the FFT display and it should still be potentially audible [after amplification, filtering...].
Which means that 24bit dithering would be technically beneficial in that case. Or, that a high performing device like that would benefit from using 32bit converters and drivers.

You can have all kind of perceptions you want. But when you share those with other people, you should be mindful of the supposed objective reality we all share. Things should make sense to others, otherwise they're not very useful in a discussion.
For context: there is still no solid proof that even 16bit dither is audible in music (see "CD vs high res" research). This doesn't prove that you definitely can't hear 24bit dither at regular levels (with noisy audio gear), but you can probably imagine why it's hard to believe.

I think it's possible, yes, albeit unlikely. That's one reason why I proposed doing that loopback recording test.

Here's an even more fool-proof version, which should make it easier to set the analog gain and prevent "cheating". It includes the same drum sample, but first peaking at -3dBFS and then again at -120dBFS.
1. Play the 32bit float and the 24bit dithered file at max volume(=0dB) in your DAW or media player (each file separately, of course).
2. Record the analog output with the gain set so that the recordings peak at about -3dBFS.
3. Remove the first (-3dB) half from the recordings, leaving only the -120dB portion that we care about.
4. Boost the volume of both recordings by ~100dB.

If your converters are top notch, the dithered version should keep more low level details from the drums and you'd hear truncation noise in the recording of the undithered version (unless there's dithering at the output of your DAW/media player or if your soundcard uses 32bit drivers).

Feel free to use your own flavor of dither instead. I used Reaper's noise shaped one for that file.
And share the recordings with us if you want.

I suggest this test: Start with your 32-bit source material (a suitable mix from your DAW). Print a 24-bit dithered copy. Print a truncated copy. Pull each into a fresh project on two separate tracks. Invert one channel, print a new mix of the difference. (Note: You can do this entirely inside the DAW, with printing file, but there are a lot of ways to go wrong and not notice.)

Now bring that in and listen by itself, using a trim plugin to boost the track to whatever level you need in order to hear it. The error should be at the lowest amplitudes if nothing went wrong, but be careful to make sure nothing went wrong. This is artificial, of course, but it lets you become familiar with the nature of the difference between the two.

Two things to consider at this point: If the output sounds like steady white noise, then you might as well give up. (Some will argue this point, but if so they have the opportunity to become famous by proving it.) And this will be the case for most files, especially ones recorded and not computed, unless you go out of your way to use synthetic instruments, digital fade, and naked reverb decays.

Once you have something that you know will have differences that are non-random and know what to listen for and when it will happen, play it back without the digital gain boost and determine if you can hear it. This will be in a dead quiet room ***. You've already said you couldn't hear -120 dB without boosting, so I think you be listening to the hiss of your amp chain and background noise of the room, but nevertheless it's a good test. And you might also be surprise that it's not so easy to even get error correlated with the source signal for 24-bit.

Variation—I like this better, but a little more subjective work: Instead of listening to the difference between the two output, listen the the differences between the original 32-bit and dithered and non-dithered. The dithered difference will sound like white noise. If the non-dithered difference also sounds like white noise, no sputtering or ripping sounds at key point, then you have no hope of telling them apart. But try anyway.

*** Yes, you're going too far to say that something too quiet to hear, when combined with something louder, becomes hearable. Evidence is the opposite ("masking")—you may be able to hear a pin drop on your desk in front of you in a quiet office, but try the test again with sound system blaring. Can you hear the pin better? Now consider the same test with something that is too quiet to hear in the first place.

If I got this right, you supplied the 32-bit float source, and the looped-back version using dither on the output. Could you also upload the same loop-back, without dither? Then we'd at least have both versions for everyone (who wishes to) to check. Thanks.

No, both files are the source to be used for a loopback test. I just pre-applied dither to one of them, for convenience.

I did perform the same loopback test myself, of course, but there was nothing down there but noise... there's definitely no point in 24bit dithering with these converters.


And maybe a better, although less "musical", test is using a constant sine wave, for example 440Hz. It's at least easier to see it on an analyzer, using a large block size and long averaging.
Here's a screenshot of it being played at -130dB and recorded through a loopback. If I boost the volume, I can just about hear it too (but keep in mind that the noise level of this device is much higher than -130dB).

I'll give that a go.

Earlevel, I don't think null testing works so well. Dither lets you hear 18dB or so I think of extra stuff that you wouldn't hear without it. A null test will just leave dither noise.

Once again, what I think people are missing is that we don't simply record something and then play it back as-is. We do a lot of processing and every time it gets rendered into fixed point (if not floating), it needs to be dithered to PREVENT truncation distortion. Failure to dither can become apparent at the end of the line even when it wasn't obvious earlier. Really flat monitors make this problem more obvious.

Why nit-pick dithering?

Because the managers and reviewers whose thumbs-up can make or break an artist's career often use very high-quality gear. They are an artist's most important audience.

First, just to be clear for everyone, since Ben was talking about a loopback test, I'm not talking about with the loopback test—didn't want that to get lost in context...

But in the box, a null test is the perfect thing to test with (at least for TPDF, RPDF, truncation). If you want to answer the question
"what is the difference between the original (32-bit float, or whatever) and the reduced word size (16- or 24-bit dithered or not)", the null test tell you exactly that.

With a properly dithered signal, the difference will always be uniform white noise, even in silent passages. I think you're saying that, and I agreed. It's still educational, for a lot of people, because it drives home the point that the only difference between the original and a dithered reduction is sample size is the noice floor.

The second part is more interesting—doing the reduction without dither. But all this is simply educational, and only a suggestion if someone want to have a better idea of what really happens (entertainment value, learning, whatever).

But if I'm misunderstanding your point on the null testing, please let me know.

Coming back to this, some caveats here. First, the 18 or 20 dB under-the-noise-floor claim is for shaped dither, where the noise is moved out of the way of the signal of interest, and for 16-bit in particular. At least that's my understanding, maybe you can hear 18 dB under TPDF, I don't know, it's not interesting enough to me to pursue because I work with either 32-bit float or 24-bit (if I have to), exclusively, and this is mainly important for 16-bit, which I haven't used for at least one or two decades. (Yes, despite Chris' claim that I think 16 bits is plenty. Utter BS.)

Lastly, you have to be able to hear the noise floor in order to have any hope of entertaining that you can hear 18 dB beneath it, at since I'm only talking about 24-bit dither, that is not going to happen. (Yeah, there will be the usual blowback at that claim, but the fact is... how many bits can you hear?

Bob, if you're talking about me (pretty likely), I reiterate what I've said many times:

First, I've said over and over—dither 24-bit—why not?

Yes, I give further information, which people can use or not use (or perhaps not use but appreciate the understanding). And I made clear it's Primarily of use if certain cases. Like deciding if a person needs to dither every send. I know you do it, I don't care if everyone does it. I won't do it, because I'm interested in making music with my time and not doing things no one can ever have a chance to hear (even with post processing).

Sure, some will say, "but I know it's, potentially at least, there, and we should always strive to make it right no matter what. OK, when we mic a guitar amp we likely have 60 Hz hum, hopefully at a level we can't hear. But if we can't hear it, do we break out a spectrum analyzer and go looking for it? Do we worry, even though we can't hear it, that post-processing will reveal it? No, we carry on making music.

It's a gear forum, and if people can claim to hear the bottom bits, I can nit-pick about what goes on down there too. Again, not telling people what to do, just giving information. Fine if people disagree with it, too. (Not fine if they make BS claims, like that I think 16-bits is all that's needed.)

And I regret that trying to get to the point in text makes it sound like I'm being emphatic. A hazard of the format. I respect what you and others on the board do.

PS—In retrospect, probably a rhetorical question and best left unanswered, but what the heck.

But if the difference between dithered audio and non-dithered is perceivably more dynamic range. Just do it at 8 bit to magnify it, then how is a null test which leaves just dither noise of any use? Clearly the difference isn't dither noise, it's the effect of it. I completely understand the argument that the effect of 24 bit dither can't be heard until post processing, like limiting is done afterwards. I'm pretty sure there is a difference I can hear there. Maybe 75% sure. If any post processing is done though, the difference definitely audible from what I remember when I tested it. 99% on that.

And yea 18dB extra perceived dynamic range was for noise shaped dither. So it's less. Have you done listening tests with 24 bit dither, especially with a post limiter maybe boosting the signal 10dB maybe? I'm surprised you can't hear a difference then. Reverbs sound more 'realistic'. Hard to put into words, but it sounds better.

If I understand you correctly: It's non-linear—you can't truncate (dithered or not) to 8-bit, and have it produce anything like a higher res version of the same that's turned to hear it. That's why you have to do exactly what you're after (24-bit dither, or not), then null and turn it up (if your goal is to hear what changed, whether their are artifact, under the hearing threshold or not). I've seen people make this mistake a lot. They truncate at 8 bits then think the horrible garbage they hear would be the same as if they truncated at 16- or 24-bit, just turned down a lot. That's not the case, not even close. Try it.

If you dither from a >24-bit source, reducing to 24-bit (or 16-bit), the difference is precisely that you've ensured a steady noise floor, just above the 24-bit level (or 16-bit level). It will otherwise sound exactly the same. This is not only mathematically trivially true, but you can test this out for yourself. If you don't want to listen to me say it, Ian Shepherd makes the point brilliantly in one of his videos. (Search for his article with video at his website, with "The truth about bit-depth and digital audio resolution", or just use that in youtube). I also have a video showing this, but enough of me, I'm happy to let Ian doing the talking.

Similarly, if you null an undithered truncation with the original source, it will reveal any potential grunge that is similarly added on top of what the original hi-res sounds like.

If you, or anyone, has a bit of source material, 32-bit float, I could not only try it, I can't return some more analysis of it, compare the dither and undithered, etc. (If I supply my own, people won't accept it, and that's understandable.)

No, I haven't tried what you mentioned, but 10 dB wouldn't be enough. The error level would still be below -130 dB.

I suspect the relevant point is that with dither or natural noise sources you can hear stuff waaaaay below that noise floor, but truncated stuff is capable of entirely removing that content and replacing it with a different, nastier noise floor.

When I say nastier noise floor it's because I found I was able to tell truncated from dithered white noise (at very low bit depths) by that unpleasant, abrasive quality. When I say 'entirely removing that content' that's pretty literal: it'll take stuff lower than the quantization threshold and simply fail to respond to it. When I say way below the noise floor, sometimes it depends what the noise floor is (for instance vinyl noise floor is largely composed of inaudible subsonic energy and briefer interfering noises in the form of crackles and extraneous noises, and different digital noise floors have different opacities, some of them being a dense midrange noise and some of them a lot brighter)

At that point it is only about 'how far down counts as 'perceived', with one camp determined to rule out anything that can't be documented ten out of ten, and another camp represented by the 'You have to live with the thing and see how well it holds up over extended use as your subconscious picks up on details that typically get by you the first time. You will end up forming opinions and learning the sound signature in such a way that you can then hear what the thing's really doing and evaluate it as such'

I've got an idea on a way to test for ability to perceive fugitive stuff, but not with audio: a visual test where you're watching for a character in a sea of scrolling text going by too fast to register, forcing you to rely on subconscious impressions. I can keep wprking on that.

The test where a mix is rendered out in float, double and long double variable sizes is here: https://www.mediafire.com/folder/8g3..._On_Buss_Depth and the audio files are probably of the most interest, but the zip files contain copies of every plugin that was recompiled to run at each variable size (such as, all internal variables have to be floats, or all internal variables have to be long double). That means if you wanted you could run your own tests or do something like stack 12 reverbs on top of each other, but with different variable types. All still floating point, with the mantissa always staying at effectively better than 24 bit at all times. If that's definitively enough you should be able to do anything with these plugins and come out with identical results no matter how much you did, but in practice you definitely can't, and it pretty easily gets up to ABX-able, provably distinct results.

You got me out of my hermit cave again to shake my fist up in the air.

Yes, every person should dither all 24 bit sends out of the DAW, including their monitor send. Done. Game over. Be a good influence please.

Regarding taking time away from music. Your posts in this thread are an exponentially greater waste of time in relation to the time it takes to modify your DAW's starting template to include 24 bit dither on every physical send. So spend 2 minutes to change your template one time for potentially hundreds of songs thereafter.

And the assumption that no one will ever hear it because you dont hear it, so why bother with it, makes me think you are trolling now. And from a self imposed high vantage point, but I got news for you. It's audible.

If you would, tell me to what bit level you can hear:

How many bits can you hear?

Note the audio player immediately beneath the video. Or just download the wav file here:

multi-level sweeps.wav

Are you in Southern California by chance? If so please come over and sit in my chair and bypass the 24 bit dither yourself. I would be surprised if you couldn't her something change in this room.

If you dont hear a change, it would likely mean your mental cup is too full of thought to let in the present moment, which is the only place where sound exists.

Learning how to let go of thought long enough for your ear to relax into the moment is very important. Some people are born with the ability, but most are not. I was not, and it took me a few years in my 20s to figure it out - that thinking and listening are mutually exclusive activities - and ive kept it up till now (48).

Also, if you come here, I will let you blind test me (with a blindfold). Only caveat is I pick the material and the dither(s). Some material is harder to identify the depth change than others, so I get to pick the mix. Vanilla TPDF is fine. No ear jarring sine sweeps!

Thanks for the offer, Robb. I'm in Torrance, a bit of a drive to San Diego. I'd like to take you up on it, don't know that I can any time soon, but I appreciate it.

Of course, the "ear jarring" [square] waves in the listening test are the point—if you can't hear something ear jarring in it's naked glory, it's hard to come up with a good argument as to how you will hear quieter signals with audio playing. Just giving my point of view on it, as you did yours.

I'd really be interested in taking one of the source music examples you'd be presenting, truncate to 24-bit and null, to see what the naked error (boosted) would sound like. Not requesting or demanding anything, just saying where my curiosity lies.