24 bit in and around, 16 bit out?

[Ethan Winer]

I am completely dumbfounded that anyone would deny that adding another 50% of bit depth resolution coming into a summed down 16bit out provides no audible benefit.

So you're saying you think the difference between (let's say) 0.001 percent distortion and 0.003 percent is audible?

I am telling you what I hear first hand.

Please take this in the gentle manner it is intended:

Lots of people think they heard lots of things. Audiophiles will insist they heard a difference (always an improvement, never worse) after they:

* Put their speaker cables up on cable elevators
* Replaced their power amplifier's stock AC power cord with a "better" wire
* Demagnetized their LP records and CDs
* Applied a green magic marker around the edges of their CDs

And so forth. They are certain they heard the improvement! But hopefully you understand that the sound didn't really change. When you understand why people are sure cable elevators improve the sound, you'll understand why your own non-blind tests are not conclusive.

I have had sessions that I have worked on in 24 bit and output to 16 and directly compared them to the same session where the source was converted to 16 before mix output and I am testifying first hand that there is an audible difference.

Now hold on a minute there pardner. Are you saying you did the "destructive" equivalent to what I described in Post #9 of this thread? Do you still have the rendered mixes?

wouldn't all evidence presented here be anecdotal?

Not if it was done properly in a controlled blind test! That's the whole point of the Meyer/Moran test, and that's why their test carries much more weight than informal non-blind comparisons.

--Ethan

 

[mixdownguru]

What "resolution"? The size of the dynamic range palate is increased, but the "resolution" does not change. It's still 6dB per bit, whether it's at -90 or -140dB.I'm not questioning what you hear, but I would like clarification on *why* you're hearing it. You have two different truncations going on at two different points in the process. Are they both post-converter simple truncations or are you outputting different word lengths from the converter? Or a combination of the two? Simply put, are there other variables creeping into the equation above and beyond simple word length that may be creating the difference you hear?

Ok, let me explain it this way. The added bits are first increasing the waveforms amplitudes resolution. If you were to take a side by side visual representation of the wave form, and give each bit segment a 1:1 ratio, then you would be right in that you'd have more headroom. However, this is just one application of the additional wordlength. But, if you were to give each segment a 2:3 ratio what you are doing there is allowing for higher resolution in the amplitude which will translate into greater dynamic resolution.

The why portion of your question is totally up for debate since this is purely a subjective issue and open to interpretation. However, I believe that the added resolution allows pre-output processing (hi-res send effects for example) to make truer or better informed representations of the source material. This I think is being represented in the summed output on final.

Just as with sample rate, where depending upon the converter design and components used, some converter may sound better at one rate whereas others sound better at another rate, but not because of the sample rate itself, but because of the converter design. Also, different plugs can sound different at different word lengths or sample rates. Could there be a similar thing happening here somewhere along the signal path, or have all these possible variables been accounted for?

G.

The short answer is yes, these were all accounted for and I do not believe the bit depth has made a difference in these examples. But, you do raise a very valid point.

 

[mixdownguru]

So you're saying you think the difference between (let's say) 0.001 percent distortion and 0.003 percent is audible?

If you wish to simplify it to that point, then I would have to answer this way, you are still talking about 3x the distortion. Now, simply put on a single render it won't make a difference - however, with this happening on on tracks 1-76, it will once summed. Not to mention the effect on processing as mentioned in my previous response.

Please take this in the gentle manner it is intended:

Lots of people think they heard lots of things. Audiophiles will insist they heard a difference (always an improvement, never worse) after they:

* Put their speaker cables up on cable elevators
* Replaced their power amplifier's stock AC power cord with a "better" wire
* Demagnetized their LP records and CDs
* Applied a green magic marker around the edges of their CDs

And so forth. They are certain they heard the improvement! But hopefully you understand that the sound didn't really change. When you understand why people are sure cable elevators improve the sound, you'll understand why your own non-blind tests are not conclusive.

I understand those situation, but they aren't applicable. You're trying to tell me that maybe the difference I hear is imagined. But if you question the validity of responses here, why even ask the questions. I don't want to go into the whole "I've done this and I've been engineering that long and blah blah blah... ". I do know a little bit about the subject and just wanted to share it.

Now hold on a minute there pardner. Are you saying you did the "destructive" equivalent to what I described in Post #9 of this thread? Do you still have the rendered mixes?

Unfortunately, not only do I not recall #9 (but I will go back and reread it after this post), but I also do not have those mixes available. I will however, recreate it with a session I'll have on Tuesday and I would be more than happy to post it for your review. So, stay tuned!



Not if it was done properly in a controlled blind test! That's the whole point of the Meyer/Moran test, and that's why their test carries much more weight than informal non-blind comparisons.

--Ethan[/QUOTE]

 

[mixdownguru]

16 vs 24

I'm sorry for beating this one in the ground, but I would like to add 2 more thoughts....

Lots of people think they heard lots of things. Audiophiles will insist they heard a difference (always an improvement, never worse)

When we first entered the digital recording phase, there was no question that there was an audible difference. However, it was once believed that the digital realm was incapable of truly reproducing audio and that analog was a truer medium. It was only after many years of debate, trial and error that we've come to understand that the difference was that it was actually truer to the source and that we were listening to colored sound from the beginning and that the "warmth" we were seeking were really distortion artifacts introduced by each component in the analog chain.
My point here is that studies aren't always proving what they think they are.

Not if it was done properly in a controlled blind test! That's the whole point of the Meyer/Moran test, and that's why their test carries much more weight than informal non-blind comparisons.

--Ethan

And, my point about all studies presented here as being anecdotal, was that unless you are there while the study is being conducted, then your reading about it, or posting about it or referencing it is still anecdotal. It was just play on semantics and the fact that we are here on a forum.

 

[SouthSIDE Glen]

But, if you were to give each segment a 2:3 ratio what you are doing there is allowing for higher resolution in the amplitude which will translate into greater dynamic resolution.

"If". The problem is, that "if" is not happening. Whether at 16 or 24 bit, each bit represents a 6dB range. A waveform of xdB dynamic range will take the same number of bits to represent at either word length. All that's happening by increasing the word length is that you are decreasing the level of the digital noise floor from -96 to -144 dB.

The why portion of your question is totally up for debate since this is purely a subjective issue and open to interpretation.

Not at all. If there are different circuits or algorithms creating the actual 16-bit data being compared in the test, the difference in properties between the two would need to be eliminated or accounted for in order to definitively say that what you are hearing is indeed because of the word lengths themselves. This is why I was asking for clarification on the actual mechanisms involved - i.e. the full signal path - in the two situations being compared.

Even the same make/model of hardware or software can inject it's own artifacts in the way of a difference in sound in how they operate at the two word lengths, not because of the word lengths themselves, but rather because of idiosyncrasies in the 'ware's design. This is more than just esoterica, because that means that what one hears will not necessarily also apply to others using different 'wares, and one's particular personal experience, even if it is true, is not enough to be able to make a blanket judgment that the word length itself is better or worse one way or another, unless they can eliminate the potential other variables from the equation.

G.

 

[mixdownguru]

16 vs 24

"If". The problem is, that "if" is not happening. Whether at 16 or 24 bit, each bit represents a 6dB range. A waveform of xdB dynamic range will take the same number of bits to represent at either word length. All that's happening by increasing the word length is that you are decreasing the level of the digital noise floor from -96 to -144 dB.

I'm afraid that's just not true. Are you saying that if my source is utilizing bit 22 on a 24 bit source, then my 16bit would be maxed at 16??? That just isn't the case. But that does bring up something else. During the conversion process, are the excess bits being truncated, quantized or dithered? BTW, each bit does not directly translate to 6db of range, this is just an average as each bit has an exponential variance.

Not at all. If there are different circuits or algorithms creating the actual 16-bit data being compared in the test, the difference in properties between the two would need to be eliminated or accounted for in order to definitively say that what you are hearing is indeed because of the word lengths themselves. This is why I was asking for clarification on the actual mechanisms involved - i.e. the full signal path - in the two situations being compared.

No, what I'm referring to is truly an in-the-box algorithmic conversion.

even if it is true, is not enough to be able to make a blanket judgment that the word length itself is better or worse one way or another

But I don't think the original question posed was which was better sounding. I believe the question being asked was if there was a difference or any benefit to the added bits.

 

[SouthSIDE Glen]

I'm afraid that's just not true. Are you saying that if my source is utilizing bit 22 on a 24 bit source, then my 16bit would be maxed at 16???

I'm not quite sure I understand your question; "16th bit maxed at 16" makes no sense to me. A couple of things to consider:

First, as has already been mentioned by at least a couple of others here, the actual dynamic range from peak to noise of the analog signal being converted is going to be - if you're very lucky, if you have A-list gear, and if you can game the gain structure well - maybe 65dB max. If you set the input gain on your converter (just for illustration) so the signal peaks at 0dBFS, the analog noise floor is going to be coming in at -65dBFS. This means that any bits of "resolution" below the 11th or 12 bit are meaningless because they will be swamped by the analog noise almost as if the analog noise is acting as it's own kind of analog "dither" (in a manner of speaking.)

The second consideration to help illustrate the first is to look at that situation in reverse, set the converter gain so the analog noise floor is brought down to the digital noise floor, down at the 23rd or 24th bit, giving you the maximum potential range and "resolution". Nothing changes except the relative digital volume, all you're doing is shifting the signal representation down so the signal is now peaking at some -73dBFS instead of 0dBFS. There is no increase in "resolution" by shifting the bits down and using all 24, because there is no change other than a linear changes in the overall volume of everything.

Third: the above two points are equally as true with 16 bit as they are with 24 bit. The only difference is the point at which the digital noise floor comes in.

Fourth: "resolution" and "accuracy" do not always equate. Adding decimal places to a decimal value or bits of "resolution" to a binary value only matters if the numbers being filled into those extra placeholders are actually accurate. For example, take a hypothetical measured value for pi of, say, 3.14159. Simply adding four more decimal places to it and making it 3.141590000 does not increase it's accuracy. Nor does it increase it's accuracy if the measurement value itself is not accurate at that resolution. A measured value of 3.141597345 is NOT more accurate than 3.14159, and in fact is *less* accurate as compared to the real value of pi, because the measurement itself loses accuracy at finer values.

Imagine for a minute a world where we can somehow pump an imaginary signal into the converter unburdened and unencumbered by upstream signal chain noise, so we do indeed have a wide dynamic range signal rivaling that of the 24-bit digital canvas. It's most likely that the actual in-studio sound levels are going to be, again for illustration, peaking at around 120dB SPL. This can be covered in 20 bits, with the last four bits being increased "resolution" of the value of the 20th bit. This means at the 24th bit, we are representing a value of 1/16th of a decibel.

Leaving aside that this is several times less less than what the average human ear can ever hear, and leaving aside the fact that the live room in which the recording is made is probably not an anechoic chamber and will have an amount of it's own self-noise that will be well abouve 1/16th dB, let's focus on the converter itself. We would need to make several assumptions regarding the "resolution" quality of that converter in order to make that 24th bit significant relevant; it would require an accuracy of twice that - 1/32nd of a decibel - in both the analog preamp circuitry of the converter and in the conversion value itself.

With a more realistic analog dynamic range of 65dB, even if we ignored the analog noise, "utilizing" all 24 bits would mean a level of "resolution" at the 24th bit of some 0.00003rd of a decibel, requiring an accuracy of twice that in the converter itself.

Assuming three one hundred thousandths of a decibel could even be heard by even the finest of ears (they can't), and assuming that it wouldn't be washed away by the noise floor anyway (it will), we'd need to assume that the converter is going to be accurate on both the analog and digital sides of it's circuitry to within some 1/100,000ths of a decibel or better to make those last bits significant.

G.

 

[ofajen]

First, as has already been mentioned by at least a couple of others here, the actual dynamic range from peak to noise of the analog signal being converted is going to be - if you're very lucky, if you have A-list gear, and if you can game the gain structure well - maybe 65dB max. If you set the input gain on your converter (just for illustration) so the signal peaks at 0dBFS, the analog noise floor is going to be coming in at -65dBFS. This means that any bits of "resolution" below the 11th or 12 bit are meaningless because they will be swamped by the analog noise almost as if the analog noise is acting as it's own kind of analog "dither" (in a manner of speaking.)

OK, Glen, I'm thinking you may be cheating me a few bits. I know that we can, and often do, easily hear and distinguish signals well below the noise floor. As someone with five young kids, this happens every day! Of course, here I'm talking about high noise levels and signals that are still quite audible, but lower. Let's say in those situations, it is not unrealistic to hear a signal that is 15 dB or maybe 20 dB below the noise floor. So, shouldn't you give me credit for about 3 more bits below the noise floor? Or does this not apply to signals below the noise floor when the noise floor itself is pretty low? Of course, if you give me three more bits, that's still only 14 or 15....


Cheers,

Otto

 

[ez_willis]

 

[mixsit]

The second consideration to help illustrate the first is to look at that situation in reverse, set the converter gain so the analog noise floor is brought down to the digital noise floor, down at the 23rd or 24th bit, giving you the maximum potential range and "resolution". Nothing changes except the relative digital volume, all you're doing is shifting the signal representation down so the signal is now peaking at some -73dBFS instead of 0dBFS. There is no increase in "resolution" by shifting the bits down and using all 24, because there is no change other than a linear changes in the overall volume of everything...G.

Which where/why the term 'resolution gets slippery. The % of noise and distortion vs signal still does go up as the level gets lower.

 

[SouthSIDE Glen]

Oops, accidental duplicate post. Mods please remove.

G.

 

[SouthSIDE Glen]

OK, Glen, I'm thinking you may be cheating me a few bits. I know that we can, and often do, easily hear and distinguish signals well below the noise floor. As someone with five young kids, this happens every day! Of course, here I'm talking about high noise levels and signals that are still quite audible, but lower. Let's say in those situations, it is not unrealistic to hear a signal that is 15 dB or maybe 20 dB below the noise floor. So, shouldn't you give me credit for about 3 more bits below the noise floor? Or does this not apply to signals below the noise floor when the noise floor itself is pretty low? Of course, if you give me three more bits, that's still only 14 or 15....

Of course that last sentence is important; it'll change the math, but not all that significantly.

But more to the main point of your post: yes you are correct that if you have, say, a fan running in the room that you may hear a pencil drop at a volume somewhat lower than the sound of the fan.

But - as I understand it - when you go to record this situation, something different happens. If that sound is captured (and preamped) at a level lower than the composite analog noise level of the downstream chain, the analog noise will in effect mask the lower level stuff because it will get buried in the circuit noise itself. By the time it gets to digital, it's too late, all the digital will do is more or less faithfully reproduce the analog noise.

G.

 

[mixsit]

I sited the nature of the 'noise at the bottom of the Lex's, and it has been a very long time since I poked around at -80 on a 16bit DA30 but as I recall in both cases it is not 'hiss down there, not a 'constant, and/or modulates with the signal.
Making the case here -no, asking would the nature of the noise somewhat dictate it's ability to be masked by the noise in the analog end?

 

[SouthSIDE Glen]

I sited the nature of the 'noise at the bottom of the Lex's, and it has been a very long time since I poked around at -80 on a 16bit DA30 but as I recall in both cases it is not 'hiss down there, not a 'constant, and/or modulates with the signal.
Making the case here -no, asking would the nature of the noise somewhat dictate it's ability to be masked by the noise in the analog end?

Two things to remember here, if I understand your post properly (correct me if I am misunderstanding). First is that (again, as I understand it) it's not so much the question of the analog noise masking the digital noise as it is the analog noise masking the low level analog signal. By the times things get to digital, the "nature" of the analog sound to be converted is already pretty much set in stone (not counting any unintentional artifacting possibly caused by the conversion process itself.)

Second is that "noise" in the digital world is not the same thing and does not necessarily mean the same thing that "noise" means in the traditional analog sense. I think this is what you're witnessing in your description. I hate traveling down this road in the forums because this is stuff that very few people in this world fully comprehend (including, admittedly, me in the majority that lack *full* understanding), and it gets to be really sticky to try and describe and discuss.

But that said, in the digital world, where we're dealing with information theory and not acoustical matters, "noise" is literally the opposite of "information". That's all. Any data that carries no useful information or that obscures or dilutes useful information is called "noise". This does not necessarily correlate to anything we would normally consider to be "noise" in the macro, analog world.

In the most basic and simple aspect, the digital "noise floor" - i.e. the last bit in a digital sample - is not a level of "hiss" in the classical sense, but rather is considered noisy because it is going to be, by definition, inaccurate. This inaccuracy is what is often referred to (perhaps too often) as quantization error; i.e. it is for an infinite amount of possible analog values to accurately represent these analog values with a limited number of digits, whether it's 16 bits, 24 bits or 1 million bits. No matter how many bits you have, that last bit will only an approximation. That approximation, that quantization error, is the "noise" in the digital signal. And it is, of course a "floor" because that's the last digit, the lat bit in the quantized value.

G.

 

[Ethan Winer]

you are still talking about 3x the distortion. Now, simply put on a single render it won't make a difference - however, with this happening on on tracks 1-76, it will once summed. Not to mention the effect on processing as mentioned in my previous response.

I already explained that too, in my Post #28. Distortion on separate tracks does not add coherently. In fact, it doesn't even add at all. The whole notion of errors from "stacking" accumulating over multiple tracks is misguided and wrong. The only thing that adds is noise. Other than soft-synths, usually the ambient room and microphone and preamp noises are far louder than the residual noise of 16-bit digital at -96 dB.

I understand those situation, but they aren't applicable.

They are directly applicable! Those people are certain they heard better sound after replacing an AC power cable etc, and they will fight you to the death in an audio forum if you tell them it's all in their imagination.

One of my favorite quotes: "Everyone understands and accepts that the placebo effect is real, but for some reason audiophiles think it never happens to them."

You're trying to tell me that maybe the difference I hear is imagined.

Well, instead of "imagined" let's just say mistaken. Human hearing is very frail. Did you ever make a mix that sounds great, only to have it sound like crap the next day? Did you ever hear someone else's commercial mix sound amazing, only to have it sound not so amazing on a different day? Did you ever tweak a snare EQ to perfection only to discover later you were actually adjusting the EQ on a muted BG vocal track?

I also do not have those mixes available. I will however, recreate it with a session I'll have on Tuesday and I would be more than happy to post it for your review. So, stay tuned!

Okay. The easiest and best way to do this is with the freeware +decimate VST plug-in I described and linked to in my Post #9.

--Ethan

 

[ez_willis]

Human hearing is very frail.

I blame my shitty mixes on that very thing.

 

[ez_willis]

I blame my shitty mixes on that very thing.

well, that and the ns10's.

 

[bozmillar]

Did you ever tweak a snare EQ to perfection only to discover later you were actually adjusting the EQ on a muted BG vocal track?
--Ethan

ha. I hate when I do that.

 

[Bristol Posse]

I love these things

They must have found the worlds most honest people and then injected them with sodium pentathol for good measure to get the no discernable difference replies in such quantities.

I've tried a few sneaky experiments by playing people identical tracks at exactly the same volume and asked for opinions and get a good quantity of fairly detailed opinions as to why A was better than B or vice versa. The very fact you are asking to compare inserts a natural subconcious bias that there is a difference that people are going to look for so to have gotten that many no difference responses is quite amazing to me.

As far as if it's different or not who cares so long as you can live with the end result. The placebo effect can work the other way too and if you believe that your recording equipment is capturing a better, more dynamic signal you may perform that tiny bit better and put your performance over the top so if it works for you then go for it.

 

[gecko zzed]

Did you ever make a mix that sounds great, only to have it sound like crap the next day? Did you ever hear someone else's commercial mix sound amazing, only to have it sound not so amazing on a different day? Did you ever tweak a snare EQ to perfection only to discover later you were actually adjusting the EQ on a muted BG vocal track?

Excellent! I've done all of those! Luckily it's only been when the client is not there, and I've been able to conceal my mortification. No one need ever know!