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

[Multicore]

Objectivity is supposed to be the keynote of this forum, is it not?

To my view, not exactly but I take your gist.

Some of you apparently get ...

It's the internet, man. What can you do? We try to learn to brush it off but that ain't always so easy. Even John Jurasek seems not entirely convincing in his claims to be unaffected here...

 

[dasdoing]

The 16-bit v/s 8-bit Blind Listening Test

Can you hear a difference? Of course... you can't.

www.audiocheck.net

 

[Multicore]

Well if the noise from dither is well below the system noise floor, which IS the case in many recordings as your typical concert hall has (significantly) more noise, then it does not matter whether you can lower the dither noise even further with more bits. The number of "usable" bits will not change.
So it depends. But in theory you are right.

Zakly! Earlier when paulg1 asserted that quiet signals get less resolution in PCM I replied: yes but that doesn't matter. Later I had a bad conscience because I didn't say why it doesn't matter. And then I thought it worth mentioning because the conditions that make it matter or not aren't trivial to establish.

And then while we're collectively engineering an arbitrary cut off, e.g. 16 bits per sample, ease and cost could come into it. For example, as Amir argued above, if there's really no downside then let's make it 20 or 24 and thereby avoid the whole argument. I like avoiding arguments, so, ok. But I'll not lift a finger to get 24-bit audio on my media server because I can't be bothered. We're all subjectivists in the end. 8))

 

[danadam]

But it is the case. "a higher bit depth recording would allow more bits for the softer sounds" is true pretty much by tautology, viz. more bits == more bits.

Sure, but it will also allow more bits for the louder sounds. The insistence only on "the softer sounds" indicates (at least to me) that something is/was not fully understood.

---

In attachment there is a run of 4 notes at -3 dBFS and then -23 dBFS, first in 8 bits then in 16 bits. There's nothing "better" about the softer sound with the higher bit-depth other than the lower noise floor. And it is exactly the same for the louder sound.

The only difference between louder and softer sounds is that the louder sound may mask higher noise floor than the softer sound might (if the noise floor is audible in the first place).

The 16-bit v/s 8-bit Blind Listening Test

Can you hear a difference? Of course... you can't.

www.audiocheck.net

Something similar but with Lady Gaga song:

How can DAC's have a SOUND SIGNATURE if they measure as transparent?

have you heard of the mystery of the vinyl renaissance? Haha that's hilarious. The more the truth talkers try to clear it up with demonstrable truths, the more the obfuscators pour smoke into the space with Russell's Teapots. Without determined pro-science moderation, it will only end with a...

www.audiosciencereview.com

 

[ElJaimito]

As the very first post said, 24 bit (or rather, 32 bit floating-point) is valuable in the recording studio where material is being processed (usually NOT classical material, but there are exceptions). Higher sample rates than Red Book are also used in the studio for similar reasons; if you are messing with mathematics on signals maintaining ample (huge!) margins is useful.

For playback, 12 bit - maybe with Dolby A 12/14 bit - is as good as the best 24 track analog recorders ever achieved (~72dB range), and all those revered releases on vinyl never achieved that after the first couple of plays. So 13 bits is plenty for playback -- as long as the material is scaled to use that range. To avoid all that faffing about, 16 bit 44.1 (or 48 for other media) is entirely adequate in all circumstances... now note that consumer Atmos is highly compressed and does not achieve that level of quality!

 

[ChrisG]

I'd just like to add a note here that 24-bit or 32-bit will absolutely be used in classical recording/capture.
In the early days of digital recording, where 8-bit or 16-bit was all that was available, you had to be careful to hit the ADCs with high levels, but not so high as to produce digital clipping. There was a narrow window between noise and distortion, and it's difficult to keep musicians in that window.


FWIW, in live environments, it's pretty routine for musicians to play 6-12dB louder going from soundcheck to show, so I typically set my input levels to peak at around -20dBFS. Of course, I use modern equipment running in 32-bit, so there isn't much of a noise penalty from throwing away a couple of bits here and there.


Chris

 

[ElJaimito]

I'd just like to add a note here that 24-bit or 32-bit will absolutely be used in classical recording/capture.
In the early days of digital recording, where 8-bit or 16-bit was all that was available, you had to be careful to hit the ADCs with high levels, but not so high as to produce digital clipping. There was a narrow window between noise and distortion, and it's difficult to keep musicians in that window.


FWIW, in live environments, it's pretty routine for musicians to play 6-12dB louder going from soundcheck to show, so I typically set my input levels to peak at around -20dBFS. Of course, I use modern equipment running in 32-bit, so there isn't much of a noise penalty from throwing away a couple of bits here and there.


Chris

I hastily add that when I said that "Classical" music might not be involved, I meant that I would not expect it to be processed through the sorts of plugins and outboard that are used for popular music. NOT that it would be recorded at anything less than 24 bit/32 bit FP! Should've been clearer! And yes, I recall the painful days of trying to consistently hit 15 bits on a 16 bit ADC...

 

[antcollinet]

But it is the case. "a higher bit depth recording would allow more bits for the softer sounds" is true pretty much by tautology, viz. more bits == more bits.

It may or may not be the case that quantization noise is well below the system/environment noise floor and/or threshold of hearing but that is an entirely different assertion. And it's a more interesting thing to teach and learn about since it raises a whole bunch of caveats, e.g. assumptions about the signals involved, mastering, and what not.

But that is not how digital audio works. More bits doesn't give more resolution - or at least not in the way most people are thinking about resolution.

Number of bits just sets the level of the noise floor. If you have a full scale signal, that is well above the noise floor. A low level signal is closer to the noise floor.

It is directly analogous (ha!!) to analogue recording - say on tape. In fact the dithered digital noise floor (if you were able to hear it) sounds similar to tape hiss.

With tape if you record a full level signal (just below tape saturation starting) it is well above the noise floor. If you record a low level signal it is closer to the noise floor. Just like digital.

the main difference is (even) 16 bit digital, if noise shaped dither is applied has a significantly lower effective noise floor than even the best analogue tape systems.

 

[D2013]

But that is now how digital audio works. More bits doesn't give more resolution - or at least not in the way most people are thinking about resolution.

Yea, sampling rate is more akin resolution than bit depth.

 

[Blumlein 88]

Yea, sampling rate is more akin resolution than bit depth.

I don't think so. I think calling high sample rate more resolution is a wrong definition. What the case is here is 16 bit with dither has a low enough noise floor there is close to no room for improvement. You can get down to near thermal noise in the electronics. So while you might squeak out more with 24 bits you'll hear no difference. This would not be true of 8 bit with dither even though 8 bit with dither is far better than most people think.

 

[danadam]

the main difference is (even) 16 bit digital, if noise shaped dither is applied has a significantly lower effective noise floor than even the best analogue tape systems.

Does it even have to be noise shaped? (I don't know what the dynamic range of the best analogue systems is).

 

[antcollinet]

Does it even have to be noise shaped? (I don't know what the dynamic range of the best analogue systems is).

The best stuido systems at the end of the era - with the best noise reduction systems were (as far as I've been able to google) pushing high 90s dB dynamic range. Perhaps even low 100s

Having said that the vast majority of RTR tape was down in the region of 70dB. Cassette - maybe mid 50s dB on a good day.

 

[Multicore]

But that is now how digital audio works.

yeahahno. My interest is how most effectively to exploit the teachable moment.

More bits doesn't give more resolution - or at least not in the way most people are thinking about resolution.

Number of bits just sets the level of the noise floor. If you have a full scale signal, that is well above the noise floor. A low level signal is closer to the noise floor.

You skipped a step: calibration, or level setting.

"Resolution" in the sense that I imagine paulg1 was imagining it -- stair steps or preferably, we hope, lollipops -- is quantization which gives us the quantization noise level relative to FS. Now, if FS is appropriately set in the playback system and environment for the human audience then the difference between 16 and 24 bit noise probably won't be audible. That's what I meant when I said (above) that the low resolution of quiet signals "doesn't matter".

Even here I'm making some nontrivial assumptions. I'm hoping to glide past "appropriately set" and "probably won't be audible" without having to elaborate and I didn't even mention the signals, are they musical, acoustic, synthetic, how were they mastered?

Let's say I record Cage 4'33" in a super quiet environment with the quietest mics, amps and converters. I include that on a release with some other tracks of normal loudness. As producer of this recording I want the technical noise to be minimized and the performance to be very serious and the mastering to be realistic, so the peak for this track on the release will be very low. But I anticipate the listener might turn the gain all the way up to hear the pianist's breathing, movements, perhaps even the heart beat -- it's 4'33" so who knows what you might find. Might want this to be a 24-bit release to allow listeners to turn it way up and get that "nice resolution" of the very quiet performance, i.e. to not have to listen to 16-bit quantization noise with the playback gain set very high gain.

Questions about what dB SPL corresponds to dBFS in playback and how the track is mastered come into this. My previous remark about riding the volume control to listen to the quiet bits louder was all about this, just not as explicit.

I think this kind of thing is interesting, don't you?

 

[ChrisG]

I hastily add that when I said that "Classical" music might not be involved, I meant that I would not expect it to be processed through the sorts of plugins and outboard that are used for popular music. NOT that it would be recorded at anything less than 24 bit/32 bit FP! Should've been clearer! And yes, I recall the painful days of trying to consistently hit 15 bits on a 16 bit ADC...

Ah, you mean how the recording is then processed. In that case, agreed: classical is generally treated with a light touch. Thank goodness.

More heavily-processed music might have 2-5 layers of processing (EQ, compressor(s), de-esser, FX etc etc etc) per channel, subgroup processing, master buss processing, etc etc etc. It's amazing that a particular voice is even recognisable!



To answer the question of this thread more directly, I can say with certainty: it is possible to create a situation where 24-bits of dynamic range will occur acoustically. You need a quiet room and a loud instrument - trivial. You do not want to listen to it.

I feel like there's a disconnect between the theoretical digital-audio-discussion and the real-world problem of acoustic signal-to-noise ratios.

 

[Blumlein 88]

Okay, first is the 32 bit float version of a -115 dbFS tone at 2500 hz. I saved it to 16 bit 44.1khz with shaped dither. You can see the tone and the resulting shaped noise floor. Do be cognizant of FFT gain. You can see the noise floor is lowered especially in the sensitive to our ears 3-5 khz range while it is allowed to stack up at higher frequencies we are not able to hear very well.

This smaller FFT size is quite close to how the noise floor would seem to our ears.

 

[Multicore]

I feel like there's a disconnect between the theoretical digital-audio-discussion and the real-world problem of acoustic signal-to-noise ratios.

Now that's where the rubber hits the road! Thank you.

The hand off between the objective domain, which includes scientific theory and measurement, and real life remains in the realm of real world practicalities, i.e. engineering and philosophy.

 

[D2013]

I don't think so. I think calling high sample rate more resolution is a wrong definition. What the case is here is 16 bit with dither has a low enough noise floor there is close to no room for improvement. You can get down to near thermal noise in the electronics. So while you might squeak out more with 24 bits you'll hear no difference. This would not be true of 8 bit with dither even though 8 bit with dither is far better than most people think.

I'm not referring to the same thing as the op.

Take a look at this and you will see what i mean.

Nyquist frequency - Wikipedia

en.wikipedia.org

 

[Blumlein 88]

I'm not referring to the same thing as the op.

Take a look at this and you will see what i mean.

Nyquist frequency - Wikipedia

en.wikipedia.org

I don't follow. Higher sample rate means more bandwidth. More bandwidth is not higher resolution. Higher bit depth might be until you have enough nothing is to be gained by more. Maybe describing an example would help me follow what you are thinking in this regard.

 

[mocenigo]

It's a good moment to meditate on this chart from Amir's deck for AES.

View attachment 416742

All the fussing over and debating the just-measurable microscopic differences in the blue and green is kinda like our version of listening to the differences between power cables, except with an AP instead of ears.

Having sources and preamplifiers that have a SINAD higher than 96dB is useful because equipment downstream will amplify signal AND noise and the latter may become audible if you have a power amp with a lot of gain and sensitive speakers. But to carry the digital content, 44.1/16 is fine. These are actually two very different matters.

 

[Multicore]

Having sources and preamplifiers that have a SINAD higher than 96dB is useful because equipment downstream will amplify signal AND noise and the latter may become audible if you have a power amp with a lot of gain and sensitive speakers. But to carry the digital content, 44.1/16 is fine. These are actually two very different matters.

How does that relate to my post?