[bigshot]

But, even though you would probably use it, 24bits would be a complete waste to record acoustic records and machines. The total dynamic range is very small, strongly limited by a high noise floor..

There would be no point to 24 bits if you used electrical transcription. But if you were miking an acoustic phonograph, you might actually need it. As I mentioned before, the machine itself expands the dynamics in the recording. My Victrola produces ear splitting volumes with Caruso records with a very low noise floor. I think that has to do with the way the sound box and mica diaphragm react to energy in certain frequency ranges. In Caruso's high Cs it amplifies it by ringing in a very loud and natural sounding way. In the higher frequencies where surface noise occurs, it attenuates it. The horn also amplifies and has particular resonances. These things effectively expand the dynamic range beyond the level encoded in the grooves of the record.

Is that the different pre-emphasis applied to early records before the RIAA standard was adopted?

This is even before electricity. The same companies that made phonographs also sold records... Victor made Victrollas and sold Victor records, Columbia made machines and records, the same with Brunswick and Edison. Each company had a sound lab whose job it was to design the recording equipment to suit the machines they made, and vice versa. Some companies' records could be only played on their own brand of machine, like Edison and Pathe. You couldn't even play an Edison or Pathe disk on a Victor phonograph.

The music was recorded by performing into a large horn which funneled into a cutting lathe that cut the grooves in a wax master. A clockwork motor ran the lathe. The records were played back with the same process turned around backwards, a clockwork turntable, the needle and soundbox tracked the grooves and funneled the vibrations out through a horn which amplified the sound. The composition and configuration of the soundbox and diaphragm and the shape of the horn and the cutting lathe were all designed to complement each other. Very primitive technology, but very direct- vibrations being made into a physical record then played back the same way. The effect is startlingly present when you hear a really good recording and phonograph.

Here is what a recording session looked like. The horn had a limited range. Everything over ten feet away would fade into nothing. So the band had to crowd around the horn. With an orchestra, they would put musicians on swings suspended from the ceiling so they could get everyone in close enough.

To the left of this photo was the booth with the cutting lathe in it. Note the violin on the right with the horn attached to it. That's called a Stroh Violin.

I used to take a suitcase phono out to the patio of a local Starbucks and play records on weekends. High school kids would come up to me amazed. They couldn't believe there was no power plug or batteries. A lot of them had never even seen a phonograph record before, much less hear one played.

There are aspects to acoustic reproduction that could inform loudspeaker systems. They really understood acoustics and how a room affects the sound of music. Phonograph dealers would suggest putting a phonograph in the corner of a room so the walls and floors would act as extensions to the horn. This lowered bass response and increased volume. The horn made sound extremely directional which projected an aural image of the musician a few feet in front of the machine. It's a really creepy effect. They also recorded dry with no room ambience because they expected the room you were playing the record in to add its own reflections to the sound. Quite different from the way we design sound equipment now.

Once electrical recording with microphones was developed in the early 20s, manufacturers continued to optimize their records for their own brand of machine. Each company had its own compensation curve. A Victor might require a different playback curve than a Columbia. However, it wasn't consistently applied. Something recorded in New York would sound different than something in Chicago. And curves might change from session to session even. Phonograph folks get really good at EQing by ear. There are preamps with presets, but they don't come close to hitting all the variations in playback curves.

 

[pinnahertz]

There would be no point to 24 bits if you used electrical transcription. But if you were miking an acoustic phonograph, you might actually need it. As I mentioned before, the machine itself expands the dynamics in the recording. My Victrola produces ear splitting volumes with Caruso records with a very low noise floor. I think that has to do with the way the sound box and mica diaphragm react to energy in certain frequency ranges. In Caruso's high Cs it amplifies it by ringing in a very loud and natural sounding way. In the higher frequencies where surface noise occurs, it attenuates it. The horn also amplifies and has particular resonances. These things effectively expand the dynamic range beyond the level encoded in the grooves of the record.

There is no possible way you need 144dB theoretical, 120dB practical, to capture a acoustic recording no matter how its played. Do we now have to circle back around and talk about room noise, HVAC, mic self noise, none of which comes even close to requiring 24bits, and then go 40-50dB above all of that to look at shellac surface noise? Really?

Got your trusty SPL meter handy? Set it to "fast", and check how loud Caruso's C really is, then measure the surface noise in the lead-in/lead-out, and let's find out if you even need 16 bits.

 

[bigshot]

I suppose if you set your levels carefully before recording, you could easily fit it into 16 bit, but the loud peaks in phonograph records sometimes come out of nowhere. Since it was all acoustic, it was a lot more tolerant of overdriving peaks than electrical recordings. I've recorded my own phonograph several times where a peak would suddenly overdrive and turn to digital mush. I've learned to set my levels quite low to allow for it. Different records are cut at different volume levels too. Deeper wider grooves are louder than shallower narrower grooves. There was no standard. It's not like setting a level for a live musician in a studio. The phonograph can be very very loud and very very quiet depending on the record you play. There's no volume control on a phonograph. You swap different gauges of needles for different volume levels... loud tone needles, soft tone needles, spearpoints, medium tone... It all depends. The records themselves have a very narrow dynamic range, but played back acoustically, there can be a huge variation in level.

If I had to guess, I would bet that Caruso's high C would be somewhere above 120 dB with a loud tone needle. It makes my ears ring and my eyes wince sitting on the couch on the other side of the room. Close miked it would be very big. A very quiet surface with a soft tone needle would probably be around 40dB. Surface noise isn't really a big issue with acoustic playback, just with electrical transcription.

 

[pinnahertz]

I suppose if you set your levels carefully before recording, you could easily fit it into 16 bit, but the loud peaks in phonograph records sometimes come out of nowhere. Since it was all acoustic, it was a lot more tolerant of overdriving peaks than electrical recordings. I've recorded my own phonograph several times where a peak would suddenly overdrive and turn to digital mush. I've learned to set my levels quite low to allow for it. Different records are cut at different volume levels too. Deeper wider grooves are louder than shallower narrower grooves. There was no standard. It's not like setting a level for a live musician in a studio. The phonograph can be very very loud and very very quiet depending on the record you play. There's no volume control on a phonograph. You swap different gauges of needles for different volume levels... loud tone needles, soft tone needles, spearpoints, medium tone... It all depends. The records themselves have a very narrow dynamic range, but played back acoustically, there can be a huge variation in level.

If I had to guess, I would bet that Caruso's high C would be somewhere above 120 dB with a loud tone needle. It makes my ears ring and my eyes wince sitting on the couch on the other side of the room. Close miked it would be very big.

So, it's the resonances that get you?

And I don't mean to misinterpret, but it sounds like you're saying that it's harder to set a recording level to record an acoustic recording (where you can adjust and do an identical Take 2) than it is to set a level for a live performance where the highest peak will not be known until it's gone forever?.

 

[bigshot]

I think the thing that makes the Caruso peaks so big is the sheer power of his voice, the fact that the engineer cutting the wax master was expecting it and allowed for a very large groove, and the effect of the diaphragm ringing at a resonant frequency. The horn amplification would be pretty consistent across all frequencies, but there are certain frequencies that really blast. Record collectors who do electrical transcriptions call these "wolf tones" and they try to filter them out. But they're a big part of how the records were supposed to sound, and the acoustic playback system makes them sound very natural. It's not accurate, but it makes the limited abilities of the acoustic recording process sound better than without them. I wish I knew more about their theories, but all these people are dead now and a lot of their research was proprietary. It might not exist any more. Electrical recording is better documented because that was developed at Bell Labs.

Usually when I pull out my phonograph to record it, I set up the mikes, start rolling and then play a stack of records one after another. I could set levels individually for each record by monitoring it through one full playback, but that would take three times as long. Since the phonograph requires so much attention, winding and changing needles with each record side played, it's a lot easier to record with a one size fits all setting.

 

[Cutestudio]

1. Because you are NOT looking at a waveform, you are looking at a graphical representation of the digital data. Zoom in to that representation and you'll see the discrete quantisation values, the "stair step" image BUT that's just the encoded digital data not the decoded data that come out of the DAC. Once decoded, there is no "stair step", a continuous analogue waveform!

You're not telling me anything new here, I'm a bit puzzled at the point you are trying to make. All you are saying is digital is digital and analog is analog, didn't we realise that 30 years ago? Welcome to 2017 BTW.

2. It's not a nice or nasty noise, it's white noise and, it's way below the noise floor of the recording/reproduction with 16bit.

It must be a PWM noise because the points between don't exist, whereas they do exist in analog white noise.
Have a listen to some quiet 8bit dithered music, it's a much rougher sounding hiss than analog white noise.
As for 16bits being 'way below the noise floor', can you perhaps explain why people can hear the difference between different dither methods? If 16 bits is good enough, shouldn't the shape of the dither noise be irrelevant?

The result of dithering is a perfect, error free signal plus some benign white noise!

Oh dear, a triumph of belief over reality.
Dither is simply a way to disperse quantisation errors into the digital stream in the long term. You seem to forget that the quantisation still exists and is still distortion:
For low/mid frequencies and continuous tones of course dither works, we all know that, but it simply can't correct the quantisation errors of short transient events.
Go look at a real waveform of a click or something quiet with some bite. Zoom in and have a look at your quantisation steps for the quiet sharp transients, note that compared to the analog some of these vertical points are - by definition - in the wrong place, creating the wrong shape.
Your dither doesn't correct them because there is no way it possibly can.

Remember: low/mid frequencies and continuous (periodic) tones. Have you got it yet?
No one has ever claimed it works for short transients except perhaps you, and you're simply wrong, sorry.

As you seem completely oblivious to what dither is perhaps some more study for you is required?
https://en.wikipedia.org/wiki/Dither#Usage

Dither should be added to any low-amplitude or highly periodic signal before any quantization or re-quantization process, in order to de-correlate the quantization noise from the input signal and to prevent non-linear behavior (distortion).

There are many other such references, try Google, knock yourself out. They all say the same.

24bit is very useful, just not for a consumer music distribution format.

There is no reason why it should not be a consumer format, in fact it is already common on many DVD videos, perhaps you didn't know?
Additionally 16 bit with a shaped dither is technically unsuitable for further digital processing, so that counts out digital room EQ and full digital systems with digital level and digital crossover filtering, which would be far better done with 24bit or at least TPDF dither.

Additionally there is more than one study that shows people can tell the difference between 16 and 24bit on a consistent basis which again disproves your invalid assertion that 16bit is perfect.
http://journal.frontiersin.org/article/10.3389/fpsyg.2017.00093/full
http://www.aes.org/tmpFiles/elib/20170620/18296.pdf

 

[bigshot]

If you understand how human hearing works, and you understand the difference between 16 and 24 bit, you know exactly what difference it makes. For a recording studio, it's useful. For listening to music in your home, it's not needed. But it does serve as great sales pitch to people who falsely believe it makes a difference, so they tout it in ad copy. Because they're selling it to you, it doesn't mean you really need it.

 

[WoodyLuvr]

Additionally there is more than one study that shows people can tell the difference between 16 and 24bit on a consistent basis which again disproves your invalid assertion that 16bit is perfect.
http://journal.frontiersin.org/article/10.3389/fpsyg.2017.00093/full
http://www.aes.org/tmpFiles/elib/20170620/18296.pdf

How is it physically possible that any human ear could differentiate between 16-bits (SNR 96.33 db) and 24-bits (SNR 144.49 dB)... both of these bit depths offer a dynamic range well beyond human hearing detection capability? The frequency response is exactly the same. Really isn't the increase in bit depth only lowering the noise level (noise floor) but from a point that is already well beyond human hearing/detection? Even though 24-bit is providing a theoretical ~40-50% lower noise floor can human ears actually detect that, if so how?

By the way good sir, the first test you linked doesn't exactly state nor support your argument (please help me understand how this study relates???) and the second link you provided is not working.

Here is an interesting study directly related to 16-bit vs 24-bit: http://archimago.blogspot.com/2014/06/24-bit-vs-16-bit-audio-test-part-ii.html

 

[castleofargh]

@Cutestudio seems to only see his own idea of audio signals where sound1 + sound2 = sound1 ruined with errors from sound2's amplitude added to the original signal. sound1 is the album in 24bit, sound2 is 16bit dither. oh look how horrible it is, all the amplitudes have been changed a little, those are errors as the samples aren't exactly where they were before, so the music is ruined.
and that's of course one way to look at things. not even a very wrong one IMO. but then let's take that same way to look at sound, what happens when 2 tracks are mixed? if an extra noise down at 16bit is bad enough to justify using 24bit, what a nightmare it must be to have 2 or 3 tracks mixed together, what about 5 or 10? look at all those errors and they're not down at -90dB, they're super high in amplitude. it's like completely rewriting the signal, OMG the instrument on the first track must sound horrible now. ^_^

some representations of signal are convenient and can of course tell us a lot of things, but we mustn't mistake a representation for the actual signal. we're dealing with waves, not with dots on a graph.

 

[StanD]

Here is an interesting study directly related to 16-bit vs 24-bit: http://archimago.blogspot.com/2014/06/24-bit-vs-16-bit-audio-test-part-ii.html

Good read, thanks.

 

[sonitus mirus]

How is it physically possible that any human ear could differentiate between 16-bits (SNR 96.33 db) and 24-bits (SNR 144.49 dB)... both of these bit depths offer a dynamic range well beyond human hearing detection capability? The frequency response is exactly the same. Really isn't the increase in bit depth only lowering the noise level (noise floor) but from a point that is already well beyond human hearing/detection? Even though 24-bit is providing a theoretical ~40-50% lower noise floor can human ears actually detect that, if so how?

By the way good sir, the first test you linked doesn't exactly state nor support your argument (please help me understand how this study relates???) and the second link you provided is not working.

Here is an interesting study directly related to 16-bit vs 24-bit: http://archimago.blogspot.com/2014/06/24-bit-vs-16-bit-audio-test-part-ii.html

The first link does mention the word "inaudible" in the title and several more times throughout the study. That indicates to me that the tester is not expected to hear anything that would cause this relaxed state. Can any of this "hypersonic" nonsense be consistently repeated? It seems like we are a long, long away from being able to confidently claim that ultrasounds are having any real affect on the listener, positively or negatively.

 

[WoodyLuvr]

The first link does mention the word "inaudible" in the title and several more times throughout the study. That indicates to me that the tester is not expected to hear anything that would cause this relaxed state. Can any of this "hypersonic" nonsense be consistently repeated? It seems like we are a long, long away from being able to confidently claim that ultrasounds are having any real affect on the listener, positively or negatively.

Concur, this along with this:

"... Our findings have some limitations.
First, because we used only a visual vigilance task, it is unclear whether high-resolution audio can improve performance on tasks that involve working memory and long-term memory... "
"... Second, the underlying mechanism of how inaudible high-frequency components affect EEG activities cannot be revealed by the current data... "
"... Fourth, the present study did not manipulate the sampling frequency and the bit depth of digital audio... " High-resolution audio is characterized not only by the capability of reproducing inaudible high-frequency components but also by more accurate sampling and quantization (i.e., a higher sampling frequency and a greater bit depth) as compared with low-resolution audio... "​

 

[briskly]

By the way good sir, the first test you linked doesn't exactly state nor support your argument (please help me understand how this study relates???) and the second link you provided is not working.

Second link is of the hi-res meta-analysis study that made the rounds in the hi-fi press last year. In case this sub-forum hasn't seen it enough already: http://www.aes.org/e-lib/browse.cfm?elib=18296
Not particularly familiar with the methods and limitations of meta-analysis. I do know that some of the sourced studies have particular caveats that make them poorly applicable to more real-world systems, and some have gross errors.

 

[WoodyLuvr]

Second link is of the hi-res meta-analysis study that made the rounds in the hi-fi press last year. In case this sub-forum hasn't seen it enough already: http://www.aes.org/e-lib/browse.cfm?elib=18296
Not particularly familiar with the methods and limitations of meta-analysis. I do know that some of the sourced studies have particular caveats that make them poorly applicable to more real-world systems, and some have gross errors.

Thank you for the link (that worked)... wouldn't this "extensive training" be interpreted by some as a "forced bias"?

"... Results showed a small but statistically significant ability of test subjects to discriminate high resolution content, and this effect increased dramatically when test subjects received extensive training... "​

 

[Arpiben]

@Cutestudio
I will suggest you to read "The Theory of Dithered Quantization" from R.A.Wannamaker (2003).
Since involved maths may be unfamiliar &/or complex you may read the conclusions of the main chapters.
Basically for audio applications:

• NSD Non Substractive Dither is prefered
• TPDF Triangle Probability Density Function ( Triangle shape dither) is prefered since better fullfilling error independency vs input conditions
• Dither results in analog domain can be extended to digital domain: digital dither can be
  used to feed a digital-to-analogue converter for analogue dithering applications

This should bring answers at least for your wonders regarding 'digital white noise vs analog' as well as differences between dither shapes.

Like others I did not understand your concerns regarding transients ( Nyquist ones), dither smoothing or averaging issues.
Hope you are not confusing with other dither applications, in non linear systems,for example,where white noise is used for stabilizing purposes....

Anyhow I much prefer @gregorio or @pinnahertz explanations vs some Wannamaker or Lip shitz papers
I also admire their patience/fight against misconceptions. Respect.