Is there any music that actually requires 24 bits for replay?

[Kiko1974]

You are comparing Redbook layers to DSD? I thought you were resampling DSD to redbook resampling you did yourself? There may be differences in the mastering between the two layers. Wouldn't be the first time such was the case. Use the Weiss software to resample DSD to 16 and 24 bit and compare the results in Deltawave as already suggested. Or provide a 1 minute sample of each and one of us can do it for you. You can also use that to compare to any redbook layer on the SACD to see if there is a difference in the two layers. It is quick, efficient, reliable and accurate. You can easily see if there is even a chance there is a difference you can hear much more simply than doing listening tests.

As I explained before, I rip my SACD and then extract their tracks on a Windows laptop using ISO2DSD. For my LISTENING test I converted the DSD files to 44.1/16 and 44.1/24 with Weiss Saracon. For 44.1/16 and 44.1/24 listening tets I never used the Red Book layer on my hybrid SACD's.

 

[Multicore]

Can you give me any scenario where this might be the case (assuming from the same master, of course)?

No, I cannot. My answer was "2. Nope." and I compared believing otherwise to believing in going to heaven.

You could try using the recipe described here

Is there any music that actually requires 24 bits for replay?

I'm positively sure that I can hear a difference ... And yes, this was a full blind test, What was your score? That way we don't need to guess what being "positively sure" means. How can others confirm this? If you can't share test files (e.g. copyright) can you explain how to prepare them?

www.audiosciencereview.com

 

[antcollinet]

Now just ignoring that, 24 bit is 144 db. If we use the Dolby data saying we play as loud as 129 db (which is a bit much, but surely as much as one will need) we are going to have to find a venue where the noise is -15 db SPL unweighted to use 24 bit. With similarly quiet listening rooms to match.

And electronics at least 15dB better than the best electronics on the market.

 

[Blumlein 88]

As I explained before, I rip my SACD and then extract their tracks on a Windows laptop using ISO2DSD. For my LISTENING test I converted the DSD files to 44.1/16 and 44.1/24 with Weiss Saracon. For 44.1/16 and 44.1/24 listening tets I never used the Red Book layer on my hybrid SACD's.

That is what I thought was the case, but you had mentioned a redbook layer so I wanted to be clear.

 

[danadam]

If the question is "... that requires all 24 bits for replay" then I think it's safe to say that no, there isn't.

If the question meaning is merely "more than 16 bits", then there are recordings (one linked below) where you can hear a difference between 24-bit and 16-bit with flat dither, in noise in quiet passages and if you listen at high enough volume. So still a few caveats

God of SINAD vs. reality we get from most available music files

How do you change the minimum when displaying vertical axis in dB? Default for me was -60dB and I can't figure out how to change it. Probably an obvious thing but I can't find the setting. Under Preferences, change the meter range. Hitting the drop down gives several choices.

www.audiosciencereview.com

For me I know that I never listen to music at volume high enough to hear 16-bit flat dither on its own, so dither together with recording's noise or dither with shaped noise is nothing to worry about.

 

[amirm]

I understand that it might be useful sticking to 24 bits in the studio (for further processing), but what's the point of using 24-bit tracks for replay?

According to MQA which has the right tools to determine the true bit depth of music, in the process of encoding their MQA library, they found music that had 18 bits of dynamic range. So more than 16.

Here is a CD player I tested as far as dynamic range:

Unless you use noise shaping in the process of conversion from 24 bits to 16 (which likely few producers use), playback above 92 dBSPL at mid frequencies will have audible noise during silent passages. The more channels you have, the more the noise will add up.

Further, if you are going to apply EQ during playback -- which everyone should be doing -- you could lose even more dynamic range.

Finally, starting with 24 bit samples means less opportunity for precision loss in the signal processing.

All of this may add up to no audible difference or small ones. It costs nothing these days to get 24 bit music distributed and stored so no reason to try hard to settle for 16.

 

[amirm]

Mahler 6. Let us give thanks we don't have to endure this every day.

View attachment 416744

What the heck is he hitting with that giant mallet??? Seems like he is aiming for that drum player!

 

[RayDunzl]

What the heck is he hitting with that giant mallet???

Dropping the hammer in Mahler’s Sixth | Chicago Symphony Orchestra

In Mahler’s Symphony No. 6, which the Chicago Symphony Orchestra will perform April 21-24 under Jaap van Zweden, the expression "dropping the hammer" takes on a literal meaning.

cso.org

 

[jsrtheta]

I went to my local used record store yesterday and most of the shoppers were teens, and twenty something, buying records not CD's. Whether there after the cool album covers or, "chasing sound" I don't know.

They are being told that LP is "truer" sound than CD.

Many of them believe it. Many of them have crap equipment, too.

 

[BeeKay]

How about infinite SQ isn’t always as important to some but the handling of the devices is? Retro hypes in our (sometimes boring) digital age seem quite logical to me. I full well know that LPs can’t compete with the cheapest of DACs. Do I care when I’m in the mood for a LP? Rather not …

 

[Frank Dernie]

According to MQA which has the right tools to determine the true bit depth of music, in the process of encoding their MQA library, they found music that had 18 bits of dynamic range. So more than 16.

I do not find this plausible.
Maybe if comparing the loudest bit of music with the silence in the studio between takes, but the music itself? Maybe a tiny handfull of big classical pieces but even then I doubt it and I prefer Mahler and Bruckner to pop music and have been taking a sound level meter to concerts for over 20 years.

Even if it were so on the master no sane record company would release any recordings with that much dynamic range because none of their customers has equipment that could play it.

IMO

 

[krabapple]

You don't need 24 bit depth for that, you only need dither.

You dither when converting down.

You do recording and production at 24 or 32 bit, then at the final stage you dither when going to 16.

You don't truncate.


Drs. Lipshitz and Vanderkooy figured this out decades ago.

 

[pablolie]

According to MQA which has the right tools to determine the true bit depth of music, in the process of encoding their MQA library, they found music that had 18 bits of dynamic range. So more than 16.

Here is a CD player I tested as far as dynamic range:

Unless you use noise shaping in the process of conversion from 24 bits to 16 (which likely few producers use), playback above 92 dBSPL at mid frequencies will have audible noise during silent passages. The more channels you have, the more the noise will add up.

Further, if you are going to apply EQ during playback -- which everyone should be doing -- you could lose even more dynamic range.

Finally, starting with 24 bit samples means less opportunity for precision loss in the signal processing.

All of this may add up to no audible difference or small ones. It costs nothing these days to get 24 bit music distributed and stored so no reason to try hard to settle for 16.

That I do believe. Whether we can hear it is a different question. I read repeatedly there is no way no human, nowhere, no matter what equipment, can hear more than 20 bits. For archival and processing purposes, it is a useful tool. As for myself, 16 bits is probably more than plenty. There are a few recordings I am intimately familiar with (and which are very well recorded) where I can tell a difference between 256 and maybe 320 MP3 and lossless, but with >90% of my extensive music collection I probably could not and nor would I care. :-D

Between 16 and 20 or 24 bit FLAC files? I know I can't and I have never seen any credible study published that shows that... other than the usual crowd that declares themselves magically golden eared and claims to hear that in their confined private space without ever proving it publicly.

 

[antcollinet]

Whether we can hear it is a different question.

And whether we care - even if we can - is yet another question.

This whole thread is deep in the forest of pins with angels dancing them, while trying to count both angels and pins to work out the average. At least as far as reproduction is concerned.

 

[Tell]

You dither when converting down.

You do recording and production at 24 or 32 bit, then at the final stage you dither when going to 16.

You don't truncate.


Drs. Lipshitz and Vanderkooy figured this out decades ago.

Yes but you said we need 24bits to avoid quantization errors becoming audible, and yeah sure it's inaudible at 24 bits since the errors is at -144dB below full scale which no human could ever hear, but then the recording will also most likely have some inherent noise which acts just like dither so the errors won't even be there.
And if you truncate down to 16bit or even lower you ofcourse eliminate the dither making it completely inaudible. But tbh the audibility of quantization errors at 16bit without dither is also probably way below of what humans can hear anyways.

So no you don't need 24bits or even 1 bit to avoid quantization errors, you just need dither

 

[antcollinet]

So no you don't need 24bits or even 1 bit to avoid quantization errors, you just need dither

That still doesn't pan out. While you are processing audio - every single calculation results in a new quantisation.

Take a sample (exact interger). Do maths on it (even as simple as a volume control). Result of that calculation is almost certainly not an exact integer - so when you force the result into an exact integer that is a new quantisation error - and it is not dithered. Then as part of DSP you are doing a whole chain of calculations, requantising each time. If you are doing DSP room correction, or applying a bunch of effects during mastering, that could be hundreds of individual calculations.

If you are doing that processing in 16 bits those repeated quantisation errors (which are essentailly truncations) absolutely can build up to something audible.

 

[MatrixS2000]

That still doesn't pan out. While you are processing audio - every single calculation results in a new quantisation.

Take a sample (exact interger). Do maths on it (even as simple as a volume control). Result of that calculation is almost certainly not an exact integer - so when you force the result into an exact integer that is a new quantisation error - and it is not dithered. Then as part of DSP you are doing a whole chain of calculations, requantising each time. If you are doing DSP room correction, or applying a bunch of effects during mastering, that could be hundreds of individual calculations.

If you are doing that processing in 16 bits those repeated quantisation errors (which are essentailly truncations) absolutely can build up to something audible.

Assuming you are not talking about doing these changes in a DAW as it would immediately flip to floating. Also you should only quantize at the end of all those manipulations not at every step.

 

[Chr1]

Still not particularly clear on this.

Playback only. Room EQ, DSP and digital volume attenuation...
16bit...24bit.

Possible audible differences?
Or, no?

 

[MatrixS2000]

^ You need to research how your DAC handles this. I use an RME so I can describe what it does. It will take a 16 or 24 bit input and convert that to 42 bit to apply all the DSP operations (EQ, volume control, etc), then down convert back to the source bit depth. This bit level increase is done to get the needed additional headroom, to do those changes losslessly.

 

[Tell]

That still doesn't pan out. While you are processing audio - every single calculation results in a new quantisation.

Take a sample (exact interger). Do maths on it (even as simple as a volume control). Result of that calculation is almost certainly not an exact integer - so when you force the result into an exact integer that is a new quantisation error - and it is not dithered. Then as part of DSP you are doing a whole chain of calculations, requantising each time. If you are doing DSP room correction, or applying a bunch of effects during mastering, that could be hundreds of individual calculations.

If you are doing that processing in 16 bits those repeated quantisation errors (which are essentailly truncations) absolutely can build up to something audible.

If you're working in any normal DAW today you'll at least be working in 32bit float which is more than enough to not have to bother, and at a MiniDSP 2x4HD (and other DSPs as well I hope?) does the same so no worries about that either. And if you're only down sampling a file then whatever software you're using will most probably dither it for you.

Still not particularly clear on this.

Playback only. Room EQ, DSP and digital volume attenuation...
16bit...24bit.

Possible audible differences?
Or, no?

16bit is probably enough, but since most DACs today works at 24bit then just go for that. No need going over 16bit though for the files or stream you're playing.