The truth about bit-depth (and digital audio ‘resolution’)

How to do this, that and the other. Share, learn, teach. How did X do that? How can I sound like Y?
Post Reply New Topic
RELATED
PRODUCTS

Post

Mastering engineer Ian Shepherd has put up an interesting video about bit-depth in audio files. He says:

"Higher bit-depth files sound better because they have better resolution, right ? 24-bit records the signal with more accuracy so it sounds better than 16-bit, and so on ?
Wrong.
In fact, the only difference between 24 and 16-bit audio is the noise floor. IF you’re doing it right."

If this seems counterintuitive, watch the whole video and be convinced.

http://productionadvice.co.uk/bit-depth ... qSjnHNU.YV

Post

If somebody think that jumping 256 levels of accuracy in representing the analog - is all the same to ears. Then you need to have your ears examined or haven't spent enough time doing mixes.

There is a difference in favour of 24-bit.

If you have -96dB or -144dB theoretical noise floor, that does not matter though. If to accept lowest level is actuall +/- one step is a signal - that is -90 dB and -138 dB, but sound cards I've seen usually have -120 dB specs if actually 24-bit, but fact is 20 bits is most common among A/D-converters(144-24=120 dB).

Then if you have a preamp with better than 80 dB noise floor that's good. Line level usually do better though.

To even have surrounding more quiet than that is difficult. I turned off fridge, freezer and closed all doors - and while no traffic outside - and a mike in the room and gain on preamp full 60 dB was just below -74 dB or something.

Quite a few sample libraries are recorded 16-bit - so you won't gain running that in 24-bit - true.

But if you record anything analog and have 24-bit sample libraries - you certainly are better off.
And there is a reason formats for 24/96 like SACD was created, and recording studios also do that to a large extent. If you do 100 tracks, and every track has that tiny little bit better quality - it makes big difference to a full mix.

So theory is one thing - and reality another.

So that you would actually recreate lost information by adding dither noise - that is just not the case. You limit added quantization noise reducing bit depth by adding dither noise, introducing a random thing to the levels you lost - and since our brain is excellent in finding patterns it perceive better quality not finding these pattern anymore.

The tools he used, has fooled him - and he is just trying to create a buzz.

One thing he did not mention - was he actually recording at 192 Khz 8 bit?
That compensate quite a bit in doing a digital representation of analog material.

He said "high enough sample rate" or similar.

If you try to fit squares/rectangels to a complex analog signal - you can compensate making thinner rectangles each samples, or you can go other way and have wider rectangles and more steps.

You don't recreate original information not there anymore by adding noise - that is just .....

Post

@LFM,

That was a very useful and interesting short lesson.

Post

I'm not an expert by any means, and definitely have spent no time trying to compare the sound of 16 vs 24 bits. But I do know computer stuff and I see a lot confusion on the technical specs. I think I can share some perspective.

In 16 bit space, the dynamic range is 65k positions. For illustration, picture trying to set your speakers to play at half volume in a system that goes to 65,000. So you set it to 32,500.

What if you wanted it louder. If you set it to 32,501, can you hear it get louder?
So in specs and numbers we know it is louder now, but how much do you actually have to crank it to hear it get louder? 10 points, 100 points, 1000 points?
Say we go with 1000 to 33,500. I think you'd be able to hear that something moved.

In a 24 bit space the dynamic range is 16.8 million positions. To set your speakers to half volume you'd set it to 8,400,000. Now let's crank it up a by 1000 points like we did in 16bits. Can you hear that difference?
How much would you have to crank it to hear it in 24 bits? The equivalent to 1,000 points at 16 bits would be ~258,000 points in the 24 bit space.

So there's a TON more detail in there, and the numbers show something is definitely happening at 24 bits. The question is, can you hear it?

If it was just flat volume we wouldn't be able to hear it at all. But it's more complicated.

This dynamic range has to represent that top and bottom spots of every cycle of every wave in the music. Does it matter if your guitar wave starts off at 32,501 @ 16bits or at 8,400,001 @ 24 bits? Think about that wave going up and down, then reducing in size over its life - on the second cycle the wave tops at 31,703, on that 3rd cycle that wave tops at 27,416.

We've been using 16 bits systems for a while and we seem to like the music just fine. But is the 16 bit resolution detailed enough that adding more detail on every recording, every track, would not make a difference to a song?

What happens after you put those waves through multiple calculations inside a number of plugins- multiply, divide, find exponentials, use math to apply harmonics. Then sum it with parallel bus compression, then send it along with 80 other tracks to the master, put it through more processing then export it into another format. Could you have used more resolution then?

The modern 64 bit DAW internally sums the mixer in 64 bits. The dynamic range this can handle is a humongous number that is hard to comprehend: 18,446,744,073,709,551,616 positions.

The point is to make sure that whatever is happening with summing, virtual instruments, effects, after all the math and calculations are done, your music will not be compromised by lack of resolution.

There's probably little to no practical difference between 64 bit and 32 bit systems, both resolutions are crazy deep. But 16 bits is nowhere near that level of detail at only 65k points. Nowhere near the same league. Imagine trying to draw a chart to express the difference between 65,000 and whatever that ridiculous number is at 64 bits. Suddenly 16.8 million (24 bits) doesn't sound like that much.

Post

Noobish wrote: Fri Feb 08, 2019 2:23 am @LFM,
That was a very useful and interesting short lesson.
:wink:

I'm not sure if ^ that was sarcasm, but to be sure: LFM's post contains some of the most common (and die hard) misconceptions of digital audio representation, heh. Watch something like this instead:

https://www.youtube.com/watch?v=cIQ9IXSUzuM
lfm wrote: Fri Feb 01, 2019 2:55 amIf you try to fit squares/rectangels to a complex analog signal - you can compensate making thinner rectangles each samples, or you can go other way and have wider rectangles and more steps.
That is not how it works.

Ah, also, this is a nice workshop to watch in full. I'll link the bit depth relevant moment here, but it's a nice talk all the way through, too:

https://youtu.be/BYTlN6wjcvQ?t=2748
lfm wrote: Fri Feb 01, 2019 2:55 amQuite a few sample libraries are recorded 16-bit - so you won't gain running that in 24-bit - true.
But if you record anything analog and have 24-bit sample libraries - you certainly are better off.
That is not how it works, either.

(Talking about anything even remotely current, "quite a few" sample libraries aren't recorded in 16 bit. Higher bit depth is used in the recording stage to get appropriate noise floor for the raw material -- which often comprises of both loud and quiet articulations in total. Quite a few sample libraries might be distributed using 16 bits as the target bit depth for the final product, and when done correctly, even the very quiet articulations still sound great -- as they have been leveled appropriately from the raw recordings so that they are reproduced well in 16 bits. Then they are velocity zoned, using corresponding level attenuation in the parameters described by the sampler format accordingly.

Adding to that, it's instrument/articulation dependent, and any sample library developer worth their salt will of course use their best judgement in deciding whether a particular product contains single samples that require a stellar noise floor all the same, in the distributed format as well; piano libraries with forte keys sustaining and decaying all the way through to silence comes to mind as an extreme example of a product with single samples going from a loud signal to a very quiet one :) )

And so it goes.

Not interested in debating or arguing over this, as it's been seen a million times on the net before, and it's pretty evident what usually ensues :) ... Furthermore, I know it's all well-meaning, even. I'm sure LFM just wants to educate people from the viewpoint of really believing this stuff works in that manner. I'm just reminding that the math, the well devised listening tests and the actual factual information is out there, and if you aren't sure who to believe, personally contact someone who is a professional and an expert on these matters (for example a digital audio researcher at a university in your area) and ask who to believe, haha.
Last edited by Guenon on Sun Feb 10, 2019 7:53 am, edited 1 time in total.

Post

Yeah this bit depth thing is so overrated. The question shouldn't be is 24 bit better but do we absolutely need it? I would argue easily no we don't. CD's are 16 bit and many feel that sounded too clear compared to tape.

Post

SoundPorn wrote: Sun Feb 10, 2019 7:49 am CD's are 16 bit and many feel that sounded too clear compared to tape.
Well, that is the lost information I was talking about - it's sterile and pale.
And why vinyl had a return when people heard the difference.


Theory is one thing, reality another.

If anybody like in OP example going 24-bit down to 8 - bit and all you get is more noise, nothing essential lost - believe that if you want. Every bit left in 8-bit has lost 4096 levels of information from 24-bit - it goes without saying even.

Adding dither noise while reducing bit depth, you get less artifacts - you restore nothing of what is lost.

Noise floor matters - no doubt, but that is not the full story. You loose musical information as well going down in bitdepth. So there are multiple reasons to stay in 24-bit or 32-bit float or so while recording/mixing.

The MP3-generation probably never heard vinyl. You basically have nothing above 15kHz in vinyl - but difference in sound is noticable compared to cd. The finest details are there still.

And another thing with digital overall - you have no natural limits and can go wild on EQ and compression and all. So production of digital is to blame for some of difference because there is no limit to what math can do. And the listening devices like amps and speaker cones are the same still - so there is another clash.

A mastering for 33 1/3 rpm LP vinyl has many natural limitations taken into account, you can't go wild on bass and dynamics, the carved grooves will take too much place. In DJ early days you had 45 rpm dance floor 12" single mixes for that purpose. Usually longer versions as well adapted for dance floor.
Last edited by lfm on Sun Feb 10, 2019 9:03 am, edited 2 times in total.

Post

Oh man.. I can't wait to read more bizarre misconceptions in this thread.. Just watch Monty's video and have done with it. Then go have fun making music music and stop worrying about 'sterile and pale' music - it's all in your head ;)
Mastering from £30 per track \\\
Facebook \\\ #masteredbyloz

Post

do_androids_dream wrote: Sun Feb 10, 2019 9:01 am Oh man.. I can't wait to read more bizarre misconceptions in this thread.. Just watch Monty's video and have done with it. Then go have fun making music music and stop worrying about 'sterile and pale' music - it's all in your head ;)
I feared I lost too much of my hearing in my teens in rehearsal rooms - but was not as bad as I thought. We compensated not sounding so good, with volume. ;)

I have a couple of albums both on vinyl and "attempts" to remaster for cd - and prefer vinyl. In my case Led Zeppelin III and Band of Gypsys with Hendrix. Some static pops and stuff in between, but sounds way better as vinyl. Some is poor remastering, I think, but still some is due to media. Companies mostly done cd versions to keep in circulation making money I think.

And DAC's I use are modified with the best capacitors there are on market, since standard components were not the best. Major difference. But cannot reproduce what is not there anymore on the media of course.

I combined some results this guy had:
http://lampizator.eu/LAMPIZATOR/LAMPUCE ... zator.html

I even found that running a Sony DVD player with spdif out sounded worse than running spdif out on RME from PC- I could not believe it. Tiny bit more harsch sounding.

And found reason, looking close at suggestions for changes to PLL circuit on Lampizator site. Onboard time constant for jitter rejection was 40ms instead of suggested 4ms by manufacturer for CS4397. So RME had less jitter out than this optical player from Sony. Changing those components - I could not hear difference anymore.

So my ears are good enough, I think.

He "lampzator" is selling DAC's starting at around $2000-$5000. But since having tubes in there - not so good for revealing bad mixes.

Thought about getting reference material on SACD, but they don't send digital out as is so I can use my DAC. So to compare normal cd version and SACD version is not possible on external DAC as far as I have found(some industry agreement not have hardware players do 24/96k on SPDIF out.

So thinking of getting SACD compatible player for PC running that out to compare going 24/96k to normal CD - but have to find good reference material first. Often SACD comes with CD version as well.

Post

lfm wrote:The MP3-generation probably never heard vinyl. You basically have nothing above 15kHz in vinyl - but difference in sound is noticable compared to cd. The finest details are there still.
How can you say that with dry eyes?

Firstly, the upper limit is closer to 18 kHz than 15 kHz. High frequencies are less of a problem in vinyl than low frequencies.

You acknowledge things get lost. How can these be preserved at the same time? Is this quantum mechanica or what weird world can have non-existing things existing nevertheless?
We are the KVR collective. Resistance is futile. You will be assimilated. Image
My MusicCalc is served over https!!

Post

lfm wrote: Fri Feb 01, 2019 2:55 amIf somebody think that jumping 256 levels of accuracy in representing the analog - is all the same to ears. Then you need to have your ears examined or haven't spent enough time doing mixes.
What has 8-bit got to do with anything? 16 bit audio has 65,536 possible values for each sample. If you think you can pick things at that resolution, you are delusional.
There is a difference in favour of 24-bit.
Yes, 24 bit audio has16,777,216 possible values. That's rather a lot, isn't it?
But if you record anything analog and have 24-bit sample libraries - you certainly are better off.
No, you aren't.
And there is a reason formats for 24/96 like SACD was created, and recording studios also do that to a large extent.
Yes, that's because they have a product to sell to customers and nobody ever went broke underestimating the intelligence of their customers, so they assume you'll be fooled. Seems they were right in your case.
jochicago wrote: Sat Feb 09, 2019 5:58 pmThere's probably little to no practical difference between 64 bit and 32 bit systems, both resolutions are crazy deep. But 16 bits is nowhere near that level of detail at only 65k points. Nowhere near the same league. Imagine trying to draw a chart to express the difference between 65,000 and whatever that ridiculous number is at 64 bits. Suddenly 16.8 million (24 bits) doesn't sound like that much.
The point is it doesn't matter which has more detail, what matters is how much detail you need. Imagine a bicycle gear system - mine has 24 speeds but I find that 90% of the time I change up or down at least two gears at a time, because they are too close together. So realistically, I only need 12 gears but the thing is, which ones are better for me, the odds or the evens? It's the evens in my case but for someone else it might be the odds, so having all 24 still makes sense, But what if someone came out with a 65,000 speed gearset? Would having to skip 10,000 gears at a time make sense to anyone? No, of course not because 24 is enough. And that's what we have here - 16 bit is more than enough, you should only need 24 bit or 32 bit for mixing but after you have finished mastering, reducing it to 16 bit won't affect the audible quality of the work at all.
lfm wrote: Sun Feb 10, 2019 8:54 amWell, that is the lost information I was talking about - it's sterile and pale.
And why vinyl had a return when people heard the difference.
Do you know why records sound so "warm"? It's because they have to compress the bejeesus out of it to put on vinyl. It has absolutely nothing to do with bit rate or anything else.
The MP3-generation probably never heard vinyl. You basically have nothing above 15kHz in vinyl - but difference in sound is noticable compared to cd. The finest details are there still.
I had a mate who was a nut-job audiophile. His stereo was worth more than a new car - hand built pre- and power amp, hand-lacquered speakers and about 400kg of wires running between them to minimise signal loss. He worked for Sony so one weekend he brought home a top-of-the-line CD player and we did some A-B comparisons between vinyl and CDs. Yes, on his system you could hear a difference but we both agreed that you couldn't say one sounded better than the other, just that they didn't quite sound the same. That was in about 1987, when everything was still recorded to tape (AAD) and digital audio production was in its infancy. Today we can do incredible, miraculous things that I couldn't even have dreamt of in 1987 so these days I reckon a CD would sound clearly superior to anything on vinyl.
And another thing with digital overall - you have no natural limits and can go wild on EQ and compression and all. So production of digital is to blame for some of difference because there is no limit to what math can do. And the listening devices like amps and speaker cones are the same still - so there is another clash.
I'm sorry but this is absolute garbage, Are you honestly trying to tell us that speakers are designed to handle only old-school signals? Obviously you don't appreciate what comes out of a tiny pair of Genelec monitors and what you would have needed in the 1980s to get anywhere close to the same sound quality. Amps and speakers have improved at least as much as anything else and benefitted equally from digital technology.
NOVAkILL : Legion GO, AMD Z1x, 16GB RAM, Win11 | Zoom UAC-2 | MPK Mini+ | Studio One 6.6
ARP2600, ARP Odyssey, OB-EZ, SEM, OB-1, Prestige, GR-8, JP6K, Union, Saurus,
Invader 2, Olga, TRK-01, BA-1, Thorn, Spire, VG Iron 2

Post

BONES wrote: Sun Feb 10, 2019 11:37 am
lfm wrote: Fri Feb 01, 2019 2:55 amIf somebody think that jumping 256 levels of accuracy in representing the analog - is all the same to ears. Then you need to have your ears examined or haven't spent enough time doing mixes.
What has 8-bit got to do with anything? 16 bit audio has 65,536 possible values for each sample. If you think you can pick things at that resolution, you are delusional.
Comments here are for those that had the time to watch the OP video - which according to video used 8-bit as example. And he did not mention which sample rate just 'high enough".

About vinyl - the RIAA is used to compress physical size of grooves in vinyl - and this in itself is compensated the other way when played back.

This in itself does not produce any warmth. What we hear on vinyl is just simply more of what was in recording to start with.

Post

No, you don't. It's a mechanical process with so many points where degradation can occur that it is simply not possible, even on the first listen. By the tenth listen, it will have degraded noticeably, even with a good turntable, tone-arm, cartridge and stylus. You have a sharp piece of diamond running through soft plastic, it's a recipe for disaster. OTOH, a CD will contain exactly the same digital information as the mastered source material, making it absolutely identical in every respect. With a CD, it is like copying a file from one folder on your hard-drive to another - it is a perfect copy, exactly as the artist/producer wanted it to be, and will remain that way forever.
NOVAkILL : Legion GO, AMD Z1x, 16GB RAM, Win11 | Zoom UAC-2 | MPK Mini+ | Studio One 6.6
ARP2600, ARP Odyssey, OB-EZ, SEM, OB-1, Prestige, GR-8, JP6K, Union, Saurus,
Invader 2, Olga, TRK-01, BA-1, Thorn, Spire, VG Iron 2

Post

BONES wrote: Sun Feb 10, 2019 11:38 pm With a CD, it is like copying a file from one folder on your hard-drive to another - it is a perfect copy, exactly as the artist/producer wanted it to be, and will remain that way forever.
... within the confinements of 44.1k @ 16 bits. Not to be pedantic, but going back to the overarching theme of this conversation. 44.1k/16 bits is not just arguably inferior to better formats, it is documented as audibly inferior to better formats:
http://www.aes.org/e-lib/browse.cfm?elib=18296

Even untrained ears have a shot of spotting the sound difference when put to a test and invited to really focus.

So you brought up an important point, and this part is what I keep going back to:
> as the artist/producer wanted it to be

As an artist/producer I don't want it to be 44.1k/16bits. If there's a chance someone could tell the difference (or take less enjoyment out of it even subconsciously). And even if subtle and borderline unnoticeable I know it's there, while just as easily I could use a format of the same file that meets a higher standard of demonstrable quality. I want to produce at a resolution high enough at which it truly won't matter, and IMO, from what I've seen/read, 44.1k/16Bits is not high enough to meet that standard.

On the other hand, I don't need my tracks to be 192k/32bits. I think at that point we are way past overkill. So what's the right size?

In my research, I determined that the minimum standard of nearly no compromise is 48k/24bits.

96k/24bits would be ideal and at that point nearly nobody, even well trained, is liable to hear a difference. But 48k/24bits puts little strain on your system, doesn't require unreasonably larger files, and your DAW and effects should be able to handle it comfortably. So that's what I aim for.

Now, ask me if I personally can hear a difference between high quality mp3 and cd format. :hihi:
(I guess I have to go test this now)

Post

jochicago wrote: Mon Feb 11, 2019 12:17 am
BONES wrote: Sun Feb 10, 2019 11:38 pm With a CD, it is like copying a file from one folder on your hard-drive to another - it is a perfect copy, exactly as the artist/producer wanted it to be, and will remain that way forever.
... within the confinements of 44.1k @ 16 bits. Not to be pedantic, but going back to the overarching theme of this conversation. 44.1k/16 bits is not just arguably inferior to better formats, it is documented as audibly inferior to better formats:
http://www.aes.org/e-lib/browse.cfm?elib=18296

Even untrained ears have a shot of spotting the sound difference when put to a test and invited to really focus.

So you brought up an important point, and this part is what I keep going back to:
> as the artist/producer wanted it to be

As an artist/producer I don't want it to be 44.1k/16bits. If there's a chance someone could tell the difference (or take less enjoyment out of it even subconsciously). And even if subtle and borderline unnoticeable I know it's there, while just as easily I could use a format of the same file that meets a higher standard of demonstrable quality. I want to produce at a resolution high enough at which it truly won't matter, and IMO, from what I've seen/read, 44.1k/16Bits is not high enough to meet that standard.

On the other hand, I don't need my tracks to be 192k/32bits. I think at that point we are way past overkill. So what's the right size?

In my research, I determined that the minimum standard of nearly no compromise is 48k/24bits.

96k/24bits would be ideal and at that point nearly nobody, even well trained, is liable to hear a difference. But 48k/24bits puts little strain on your system, doesn't require unreasonably larger files, and your DAW and effects should be able to handle it comfortably. So that's what I aim for.

Now, ask me if I personally can hear a difference between high quality mp3 and cd format. :hihi:
(I guess I have to go test this now)
Yeah, I think I'm going to be considerably more inclined to believe what a peer-reviewed, extensively tested/researched article from the Audio Engineering Society states than what some guy on YouTube claims. Thanks for the link, @jochicago. I hope this puts this issue to rest now and backs up @lfm's points.

Steve
Here's some of my stuff: https://soundcloud.com/shadowsoflife. If you hear something you like, I'm looking for collaborators.

Post Reply

Return to “Production Techniques”