[BackToAnalogue]

Yes I agree, a thoroughly nice chap and great communicator, I have read both those as well and his book about the Challenger disaster.

Just a pity about that rather dodgy experiment :>~

 

[Maxx134]

Hey I read this thread go back and forth ad infinitum with opposing ideologies and we need to realize there has to be a common sense middle ground, which I believe the recordings industry, that relies on real world results is correct regardless of the "why"..
Because it works..

24 bit is used for mixing and mastering ...

This is the simple direct answer posted long ago to answer this thread question topic...

Haha last thing I would be in agreement but also there is onething to consider here that has no basis on anyones viewpoint except the recording industry. .

And that is 24bit is accepted standards which are not based on any fads, or theories...

24bit is the accepted standard because it in the real world this works.
This makes a difference.
Thats the bottom line whether you like it or believe in it or not.

 

[RRod]

Hey I read this thread go back and forth ad infinitum with opposing ideologies and we need to realize there has to be a common sense middle ground, which I believe the recordings industry, that relies on real world results is correct regardless of the "why"..
Because it works..
This is the simple direct answer posted long ago to answer this thread question topic...

Haha last thing I would be in agreement but also there is onething to consider here that has no basis on anyones viewpoint except the recording industry. .

And that is 24bit is accepted standards which are not based on any fads, or theories...

24bit is the accepted standard because it in the real world this works.
This makes a difference.
Thats the bottom line whether you like it or believe in it or not.

 
24bits is of course better for recording and editing; no one debates that much at all. The issue is that companies are trying to convince people that they *need to re-buy their current musical libraries at 24bit so they can get "all" the music*, which is complete hogwash. Noticeable pain will start at 125dB, which is what you can get to with 16bit non-noise-shaped content in a 30dB listening environment (a library, basically). This is already way beyond the requirements of almost all listeners.

 

[Greenears]

You can get to any dB you want by changing the volume.... so I think this statement is not meaningful. Which does not mean it is not often repeated

 

[RRod]

You can get to any dB you want by changing the volume.... so I think this statement is not meaningful. Which does not mean it is not often repeated

 
It's a matter of dynamic range, not volume. And it's about not having to change the volume mid-track because the dynamic range is too high. Got something meaningful to add yourself now?
 
forgot… It's also about the fact that no music I know of on earth uses 24bits of dynamic range, and that no ADCs and DACs get that many effective bits anyway.

 

[bigshot]

I don't know of any music that uses all of 16 bit dynamic range. Most music is under 50dB. It isn't comfortable to listen to music that goes from too loud to too quiet. 16 bit covers all you would ever need for playing back music on a home stereo... and even that has dynamic range to spare.

 

[RRod]

  I don't know of any music that uses all of 16 bit dynamic range. Most music is under 50dB. It isn't comfortable to listen to music that goes from too loud to too quiet. 16 bit covers all you would ever need for playing back music on a home stereo... and even that has dynamic range to spare.

 
Yes, I've yet to see anything above 65dB from peak to lowest RMS (during music), and you can guess the kind of material where that happens.

 

[bigshot]

Wagner

 

[RRod]

  Wagner

 
Yes, the composer of Death Magnetic :3

 

[Greenears]

What is going on here for those that are not following the details is they are using a calculation of Noise Foor + N * 6 dB/bit approximation where N is number of quantized bits.  So for 30 dB noise floor that is 30 + 6 * 16 = 126 dB.  Sounds all well and good but try some other values and it falls apart.  For instance say you lower the volume so the peaks are around 50 dB SPL which corresponds to a normal conversation at 1m.  30 dB corresponds to a quiet room.  Well clearly you can hear a normal conversation in a quiet room, but now they imply you only need 3.5 bits to capture that 30 + 6 * 3.5 = 51 dB or so.  But if you try it in practice you will see you need 7 to 8 bits to capture speech at 50 dB SPL playback, and it gets noticeably clearer at 9 or 10 bits even with the same noise floor.  4 bits it breaks up.   So how come the quality differences are easily audible up to 10 bits and beyond with peaks at 50 dB and noise at 30dB? They just implied 10 bits requires 90 dB SPL to "hear" it which is rock concert loud. 
 
That's because quantization noise is non-linear and it affects amplitude phase and frequency equally and we are not equally sensitive to those things.  Also all these dynamic range values are long-term averages the instantaneous values can be much higher.  It's just not that simple. Yes more bits generally means more accuracy and eventually the errors will become inaudible but this equation is not really the correct way of expressing that statement at least in terms of implying it has to be jet-engine loud for you to "hear" the number of bits.  But it is seductive and it will be repeated on this and many other forums forever.  And no doubt I'll be ridiculed for having the temerity to even think about questioning  it

 

[bigshot]

Recordings of music don't often have more than 50dB of dynamic range and it's usually normalized up to near peak level. Below -50dB is either stone silence down the rest of the way to the limits of 16 bit or most likely room noise from the air conditioning in the recording venue. Go ahead and precisely reproduce the inaudible noise if you want, but you will never hear anywhere near the limit of 16 bit listening to music at a normal listening level in a home situation. 16 bit is overkill. Every aspect of redbook audio is overkill in the real world. People spend too much time on numbers on pages and not enough on sound they can actually hear.

 

[RRod]

  What is going on here for those that are not following the details is they are using a calculation of Noise Foor + N * 6 dB/bit approximation where N is number of quantized bits.  So for 30 dB noise floor that is 30 + 6 * 16 = 126 dB.  Sounds all well and good but try some other values and it falls apart.  For instance say you lower the volume so the peaks are around 50 dB SPL which corresponds to a normal conversation at 1m.  30 dB corresponds to a quiet room.  Well clearly you can hear a normal conversation in a quiet room, but now they imply you only need 3.5 bits to capture that 30 + 6 * 3.5 = 51 dB or so.  But if you try it in practice you will see you need 7 to 8 bits to capture speech at 50 dB SPL playback, and it gets noticeably clearer at 9 or 10 bits even with the same noise floor.  4 bits it breaks up.   So how come the quality differences are easily audible up to 10 bits and beyond with peaks at 50 dB and noise at 30dB? They just implied 10 bits requires 90 dB SPL to "hear" it which is rock concert loud. 
 
That's because quantization noise is non-linear and it affects amplitude phase and frequency equally and we are not equally sensitive to those things.  Also all these dynamic range values are long-term averages the instantaneous values can be much higher.  It's just not that simple. Yes more bits generally means more accuracy and eventually the errors will become inaudible but this equation is not really the correct way of expressing that statement at least in terms of implying it has to be jet-engine loud for you to "hear" the number of bits.  But it is seductive and it will be repeated on this and many other forums forever.  And no doubt I'll be ridiculed for having the temerity to even think about questioning  it

 
Yes, at the least you have to factor in the crest factor of the given waveform, with higher crests reducing the effective bits of the system, at least in terms of dynamic range. A factor of 10 (probably nearing max for music) corresponds to about a 3 "bit" difference in the rms and peak levels, so using rms as a rough proxy for loudness then you only get 13 out of your 16 bits for the dynamics. So now we're at 78dB dynamic range, which would take me from the quietest room in my house to a chainsaw.
 
You need some "breathing room" for the peaks. So indeed, we're simplifying when we say 16bits is "library to pain" and that 24bits is "library to death", since we're comparing peak ranges rather than rms-to-peak ranges. But it's also simplifying to not talk about dither and noise shaping or to use rms as a loudness proxy. So at the end of the day, we have to pick some "quick" way of saying you don't need 24bits for delivery. In the future perhaps I'll just re-post these examples that I made.

 

[castleofargh]

Hey I read this thread go back and forth ad infinitum with opposing ideologies and we need to realize there has to be a common sense middle ground, which I believe the recordings industry, that relies on real world results is correct regardless of the "why"..
Because it works..

24 bit is used for mixing and mastering ...

This is the simple direct answer posted long ago to answer this thread question topic...

Haha last thing I would be in agreement but also there is onething to consider here that has no basis on anyones viewpoint except the recording industry. .

And that is 24bit is accepted standards which are not based on any fads, or theories...

24bit is the accepted standard because it in the real world this works.
This makes a difference.
Thats the bottom line whether you like it or believe in it or not.

so if I follow you argument, I need a oven to eat bread.
 
 
 
 

 

[Greenears]

   
Yes, at the least you have to factor in the crest factor of the given waveform, with higher crests reducing the effective bits of the system, at least in terms of dynamic range. A factor of 10 (probably nearing max for music) corresponds to about a 3 "bit" difference in the rms and peak levels, so using rms as a rough proxy for loudness then you only get 13 out of your 16 bits for the dynamics. So now we're at 78dB dynamic range, which would take me from the quietest room in my house to a chainsaw.
 
You need some "breathing room" for the peaks. So indeed, we're simplifying when we say 16bits is "library to pain" and that 24bits is "library to death", since we're comparing peak ranges rather than rms-to-peak ranges. But it's also simplifying to not talk about dither and noise shaping or to use rms as a loudness proxy. So at the end of the day, we have to pick some "quick" way of saying you don't need 24bits for delivery. In the future perhaps I'll just re-post these examples that I made.

No that's not what I'm saying really.  Even if you set the peaks (or crests as you call them) to 50 dB SPL and compress (raise) the bases to 45 dB SPL so that you have an average (or RMS) of 5 dB dynamic range, you can still easily hear way more bits than either 5dB or 20 dB (50 - 30) implies.  This is the mixing up of THD+N and dynamic range and quantization that is often repeated - that is my point.
 
But it's really not worth arguing over any further.  I think my illustrated numbers will allow objective readers to realize that it is not that simple, because clearly the equation used this way does not extend to all values.  I think I've expressed my point sufficiently.

 

[RRod]

 
No that's not what I'm saying really.  Even if you set the peaks (or crests as you call them) to 50 dB SPL and compress (raise) the bases to 45 dB SPL so that you have an average (or RMS) of 5 dB dynamic range, you can still easily hear way more bits than either 5dB or 20 dB (50 - 30) implies.  This is the mixing up of THD+N and dynamic range and quantization that is often repeated - that is my point.
 
But it's really not worth arguing over any further.  I think my illustrated numbers will allow objective readers to realize that it is not that simple, because clearly the equation used this way does not extend to all values.  I think I've expressed my point sufficiently.

 
My point with crest factor was just that the actual variation in loudness we can capture for a given waveform depends on the shape of the waveform, in addition to lots of things, including our physiology (not all noise is equal to our ears, for instance). But yes, I guess we're talking past each other. Still, if 24bits are so necessary then someone can feel free to post an example of music or movie material that would need more than what 16bits + noise shaping can give. All that is needed is one counter-example.