• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Can we hear the bottom bits of 24?

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
I need to recalibrate my head, in a space where people are open to discussing technical aspects of audio and questioning whether it's really possible to hear something, a opposed to feeling insulted that anyone should question it. Not to disparage a useful discussion board, I think "GS" is close enough for those curious, then no one need ask. But this is more about checking out this space and people.

I questioned the ability to hear the bottom couple of bits of 24-bit. I'll briefly give the reason, for context, but I don't want to dwell on it: The typically way music is recorded results in a multitrack mixdown and likely involves processing in the digital domain. Even if we were able to record perfectly, if we want to produce a 24-bit resulting audio file, we need to either truncate which produces error correlated to the signal (not a good idea in general), or we add a little dither noise to then truncate, resulting in a similar error size but lack of correlation—the error broadband noise, a constant hiss. With normal dither, the means the difference between the original "perfect" signal and the 24-bit dithered output is ± 1 lsb. So that's by I say the last couple of bits—it's pointless to move up from that and question three or four bits.

For now, I won't go into physical limits of the electronics (Johnson and shot noise, etc.), or the dynamic range of the ear (not my expertise, but it seems if we set our maximum listening volume to a tolerable level, according to typical hearing info it seems like ~140 dB SPL down from that would be not easy to hear, to understate it).

So, I'm asking whether people think the lowest couple of bits of 24-bit audio can be heard, and why or why not. Related to what I've already said, such questions might be whether we can tell the difference between the original ("perfect") audio and the truncated or dithered versions, between 24-bit truncated and dithered versions, or even the naked nulls (the difference) between any of these. Or simply hear a known test signal of that level.

5-bit sample sweep tone
16-bit sample sweep tone
24-bit sample sweep tone

24-bit is the one you want to listen to, but you should familiarize yourself with the tone by hearing the 5-bit so you know what to listen for. The 16-bit is there too, for fun.

I'll link the an old article containing them, because you'll probably want to download at the least the 24-bit so you can play or view outside of your browser—the direct links might be awkward to download from, depending on browser: Perspective on dither

I just want to see what people have to say, maybe discuss.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
10,568
Likes
23,443
Location
The Neitherlands
I questioned the ability to hear the bottom couple of bits of 24-bit.

Those will drown in noise with absolute certainty. If not by the DAC itself then by the amp following it and even more so by our rather flawed hearing abilities.

24 bit format is handy for recording because you can have loads of headroom to capture peaks (just to limit/compress them later on in production).
A 24 bit DAC is also handy when using digital volume control.

Plenty of research has been done about audibility levels and with various types of jitter being used.
Besides test tones is not equal to recorded music and noise in actual recordings greatly exceeds any 24 bit errors.
 
OP
E

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
Those will drown in noise with absolute certainty. If not by the DAC itself then by the amp following it and even more so by our rather flawed hearing abilities.

24 bit format is handy for recording because you can have loads of headroom to capture peaks (just to limit/compress them later on in production).
A 24 bit DAC is also handy when using digital volume control.

Plenty of research has been done about audibility levels and with various types of jitter being used.
Besides test tones is not equal to recorded music and noise in actual recordings greatly exceeds any 24 bit errors.
Agree. And good point about the extra bits being handy for digital volume control—I made a similar point once on this board (in a discussion about whether whether to control volume at the DAC or preamp), now that you mention it.

But recording engineers will argue that even though the recorded signal may have more noise, there are valid cases when the noise is below the 24-bit level, and theoretically dither is a must. One way, potentially, is generated music, but even with typical recording, it's possible to get legit audio with noise below that level (digital fade, but also with floating point reverb fading out exponentially to nothing. (Of course, I contend this doesn't matter because it's too low to hear and the DAC will bury it in noise—you'll only hear it by cheating with digital gain.)
 

JSmith

Major Contributor
Joined
Feb 8, 2021
Messages
1,412
Likes
2,547
Location
Algol Perseus
My, my, my. Aren't you a jaded little dude.
Sorry am I missing something here... you just joking without a smiley, or?

@solderdude is a very experienced and senior member here, so if not joking... well, show us what you got. I know who my money would be on... :)



JSmith
 

kchap

Active Member
Joined
Jun 10, 2021
Messages
245
Likes
254
I need to recalibrate my head, in a space where people are open to discussing technical aspects of audio and questioning whether it's really possible to hear something, a opposed to feeling insulted that anyone should question it. Not to disparage a useful discussion board, I think "GS" is close enough for those curious, then no one need ask. But this is more about checking out this space and people.

I questioned the ability to hear the bottom couple of bits of 24-bit. I'll briefly give the reason, for context, but I don't want to dwell on it: The typically way music is recorded results in a multitrack mixdown and likely involves processing in the digital domain. Even if we were able to record perfectly, if we want to produce a 24-bit resulting audio file, we need to either truncate which produces error correlated to the signal (not a good idea in general), or we add a little dither noise to then truncate, resulting in a similar error size but lack of correlation—the error broadband noise, a constant hiss. With normal dither, the means the difference between the original "perfect" signal and the 24-bit dithered output is ± 1 lsb. So that's by I say the last couple of bits—it's pointless to move up from that and question three or four bits.

For now, I won't go into physical limits of the electronics (Johnson and shot noise, etc.), or the dynamic range of the ear (not my expertise, but it seems if we set our maximum listening volume to a tolerable level, according to typical hearing info it seems like ~140 dB SPL down from that would be not easy to hear, to understate it).

So, I'm asking whether people think the lowest couple of bits of 24-bit audio can be heard, and why or why not. Related to what I've already said, such questions might be whether we can tell the difference between the original ("perfect") audio and the truncated or dithered versions, between 24-bit truncated and dithered versions, or even the naked nulls (the difference) between any of these. Or simply hear a known test signal of that level.

5-bit sample sweep tone
16-bit sample sweep tone
24-bit sample sweep tone

24-bit is the one you want to listen to, but you should familiarize yourself with the tone by hearing the 5-bit so you know what to listen for. The 16-bit is there too, for fun.

I'll link the an old article containing them, because you'll probably want to download at the least the 24-bit so you can play or view outside of your browser—the direct links might be awkward to download from, depending on browser: Perspective on dither

I just want to see what people have to say, maybe discuss.
24 bits for linear PCM is 144dB down on FSD. Some amps are getting closer but doubt if a microphone and amp combo could anywhere near that level. I think @solderdude summed the reasons for 24 bit nicely.
 
OP
E

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
24 bits for linear PCM is 144dB down on FSD. Some amps are getting closer but doubt if a microphone and amp combo could anywhere near that level. I think @solderdude summed the reasons for 24 bit nicely.
I agree, but I'll add for the discussion that there is a fundamental limit of how close we can get, because there is a limit on the lowest noise floor possible. Since there is also a (slightly loose) limit on the highest voltage a DAC can put out, both in the sense of being realizable and in being compatible with other audio gear, that puts the last bits under that noise floor, unavoidably. (Just making an assertion, not laying the law—I'm totally open to anyone discussing why that might not be true.)

Mojo Audio has an interesting paper (not assuming anything one way or another about their gear, I don't know, but they aren't cheap so I give them props fro honesty):

The 24-Bit Delusion

Key conclusion: According to the experts who manufacture the finest DAC chips, resistors, and power regulators, there is theoretically no way to make electronics that are capable of discerning much greater than a 20-bit resolution (120dB dynamic range). Any company who claims 24-bit resolution from their DAC is simply full of [edit for forum decency: poo]. Oh they can decode 24-bits, because 24-bits does exist on the digital side, but the analog output stage in the world's best DACs are not capable of resolving much more than 20-bits of dynamic range.
 

Koeitje

Major Contributor
Joined
Oct 10, 2019
Messages
1,664
Likes
2,514
No, because even if we could theoretically hear it there is no hardware that can play it back.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
10,568
Likes
23,443
Location
The Neitherlands
The 24 bit resolution of at least SD DACs is real.
It can even be shown to exist but in real world, looking at the actual waveform coming out of the RCA/XLR it is below the real world noise floor.
When you average the noise long enough you can see 24 bit DACs are actually putting out these small level differences.
So 24 bit resolution is there but buried in noise making effective resolution to bottom in the 21 bit range (assuming normal voltage output levels).
 

beefkabob

Major Contributor
Forum Donor
Joined
Apr 18, 2019
Messages
1,120
Likes
1,371
Sorry am I missing something here... you just joking without a smiley, or?

@solderdude is a very experienced and senior member here, so if not joking... well, show us what you got. I know who my money would be on... :)



JSmith
I think the complaints about loudness wars are overblown, and I said so in a teasing way. But we digress.
 

pozz

Data Ordinator
Forum Donor
Editor
Joined
May 21, 2019
Messages
3,988
Likes
6,393
OP
E

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
The 24 bit resolution of at least SD DACs is real.
It can even be shown to exist but in real world, looking at the actual waveform coming out of the RCA/XLR it is below the real world noise floor.
When you average the noise long enough you can see 24 bit DACs are actually putting out these small level differences.
So 24 bit resolution is there but buried in noise making effective resolution to bottom in the 21 bit range (assuming normal voltage output levels).
Discussion, please. I was thinking about that before posting yesterday. Now, the benefit of averaging, wouldn't that mostly be a fix for repeating waveforms? So I think are you primarily talking about revealing the ability for the 24-bit working of the DAC to resolve the tiny steps (despite the noise that ultimately buries those differences)? That makes total sense, I suspected that might be the case.
 
OP
E

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
Yes, but I'm not talking about dither within a DAC, I'm talking about dithering before the DAC, as recording/mastering engineers do for the final product.

But to be clear, I think no listener will be able to tell whether the audio is dithered or simply truncated ahead of the DAC, because the difference is ±1 lsb. It's a moot point for typical dither, if you're truncating you might as well check the box and TPDF dither. I only dissent about the need to dither for more convoluted situations that are a distraction (dithering every external send loop, monitoring dither while mixing, etc.).
 

garbz

Member
Joined
Sep 17, 2021
Messages
75
Likes
86
Now, the benefit of averaging, wouldn't that mostly be a fix for repeating waveforms?
Yes, it's great for measurement, and processing of repeating signals (think ultrasonic, radio, or laser range-finding) not so much for music. Used extensively in processing of static recording as well (think telescopes taking pictures of stars).

There is however a problem with noise and that damn Nyquist guy rears his head again https://en.wikipedia.org/wiki/Johnson–Nyquist_noise with a theorem which states that fundamentally you get to a point where noise is a function of temperature and you can't move beyond it. That's sort of where we are already with today's precision DACs.

Personally I look forward to the audiophile world adopting high tech chillers to reduce that pesky noise so we can all bask in the glory of 24bit goodness. Just like my telescope... ;-)
 

Robin L

Major Contributor
Joined
Sep 2, 2019
Messages
3,640
Likes
4,952
Location
1 mile east of Sleater Kinney Rd
My, my, my. Aren't you a jaded little dude.
No, simply experienced. A normalized 24 bit track downsampled and dithered to 16 bits will sound exactly the same as the normalized 24 bit file. However, having well over 20 db of headroom makes postproduction so much easier, you can pull up the quiet bits, compress the dynamic peaks that won't work in a domestic environments, have the noise floor inaudible. Otherwise, high bit rates are mostly a waste of data. Not only can we not hear a full 24 bits of dynamic range, so far nothing actually reaches 24 bits on account of the self-noise of the analog output. Of course, the self-noise of human hearing is much worse than that.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
2,262
Likes
2,412
Location
Pacific Northwest
Pragmatically speaking, 16 bit is not perceptually transparent, so if you want perceptual transparency you need more bits / bigger sample/word size. The word sizes of most chips come in 8 bit increments, so the next step up is 24 bits. That gives you more than you really need, which obviates the need to quibble over exactly what minimum number of bits is perceptually transparent under all conditions. And this bit of extra "headroom" is useful for recording, level setting, and DSP.

PS: In some sense it's asking the wrong question whether we can hear the LSB in 24-bit audio. We can't, but that doesn't matter. We need more than 16 bits, and 24 is the next practical step up.
 

pozz

Data Ordinator
Forum Donor
Editor
Joined
May 21, 2019
Messages
3,988
Likes
6,393
Yes, but I'm not talking about dither within a DAC, I'm talking about dithering before the DAC, as recording/mastering engineers do for the final product.

But to be clear, I think no listener will be able to tell whether the audio is dithered or simply truncated ahead of the DAC, because the difference is ±1 lsb. It's a moot point for typical dither, if you're truncating you might as well check the box and TPDF dither. I only dissent about the need to dither for more convoluted situations that are a distraction (dithering every external send loop, monitoring dither while mixing, etc.).
Those were dithered and undithered 24bit signals for the linearity aka level accuracy test (in 0.25dB steps). @KSTR can confirm. Also, consider that listening is a low target compared to, say, archival. Once that data is corrupted there will be no simple means to recover it, even if the technology improves.

I agree it's mostly a futile task given where they start to manifest and can mostly be ignored except at final mixdown. But these processing errors along with noise are determining the low level accuracy barrier.

Incidentally, @KSTR has made listening examples of low level errors: https://www.audiosciencereview.com/...dible-with-music-signals-some-examples.20886/

Despite what, IIRC, Bob Katz wrote in his book, there's no way this stuff would be audible. Unless you're deliberately leaving 90dB of headroom and expect massive boost on playback.
 
OP
E

earlevel

Active Member
Joined
Nov 18, 2020
Messages
275
Likes
329
Personally I look forward to the audiophile world adopting high tech chillers to reduce that pesky noise so we can all bask in the glory of 24bit goodness. Just like my telescope... ;-)
Now, that last bit is tantalizing, isn't it? Seems like it should be its own thread to muse on. It seems like we might have to wait for a direct connection to the brain, a la Neuromancer. If we record anything (via mics, playing electric instruments), room temperature is unavoidable, so we're either looking at algorithmic noise reduction or generating music in the digital domain, to start with 24-bit-pure sound. Then, chilling amp (and speakers I suppose) is asking a bit much, so I guess we're going for something close to ear buds or headphones (and I guess drivers might need chilling?). Then our ears would still be ears, of course, but it's a fun thought process whether it's even possible (let alone practical) to get to that point. Um, I'm holding out for Neuromancer ;)
 
Top Bottom