keith

drewfx1

SvenArne

keith

Not applying any dither to a 16-bit output will most definately produce audible nasties... You can do that test yourself.

  Funny thing, when I was young and dumb (around 2006) my band and I did a whole album and accidentally omitted dithering when going from 24 to 16 bit. And you CAN hear the truncation artifacts, but only in the very end of the tracks when the tails are fading into absolute silence (plus you'll need headphones turned WAY UP in a silent room).  

If you can only hear the truncation if you turn up the volume specifically in order to hear it, that means you can't really hear it. The fact that it's definitely there and would be audible if you turned up the volume means you are can only hear it hypothetically.

For this reason, people shouldn't claim something is audible because they can hear it when they turn up the volume - that's cheating.

Having said that, if something is perhaps borderline audible, or audible only under very specific circumstances, it does make a certain amount of sense to think about it.

  It's not "cheating" to turn up the volume. The aliasing is there in the audible range whether or not the volume is on 1 or 11. The fact that you can't hear it wen the volume is low just means that your hearing is not quite as acute as someone else's. If the aliasing were at -130dB, then that would pracitically be inaudible for just about everybody... unless the aliasing were to be enhanced into the audible range via downstream processing (e.g., extreme compression or whatever). Which brings up another point: if you're going to allow nasties, don't allow them to be enhanced later. 

I'm sorry, but you are wrong. If you can't hear it, you can't hear it. Period. If someone else might be able to hear it doesn't mean you can - it's still entirely hypothetical, not real.

Assuming we're talking about borderline cases, you can only test audibility by doing real tests, not hypothetical ones. Saying, "If I turned up the volume, then I could hear it!" doesn't make it audible. That's simply not what the word means. Same with, "I can't hear it unless I turn it up, but someone else might have better ears than me, so therefore it's audible!". In both cases the claim of audibility is entirely hypothetical, not real. So it's flatly a false claim in those cases. Hence "cheating".

Now if you want to say that, in borderline cases, something that's inaudible to me might be audible to you or someone else, that's certainly possible. But it's a speculative argument, not a factual one.

Note that I'm not saying that it doesn't make sense to leave room for margin of error in borderline cases.

But this stuff needs to be kept in context. And these kind of audiophile myths that "something might be inaudible, but since we can do stuff to make it audible, therefore it's audible (even if we don't actually do that stuff)" need to get challenged.

keith

drewfx1


I'm sorry, but you are wrong. If you can't hear it, you can't hear it. Period. If someone else might be able to hear it doesn't mean you can - it's still entirely hypothetical, not real.

There's nothing hypothetical about it. You're trying to turn this into some sort of existential, philosophical discussion. 


There's an easy answer backed by well-known and accepted science: 


1.) The dynamic range of human hearing is known to be somewhere between 120-130dB.

2.) Aliasing in a 16-bit sample must exist w/ in the 96dB dynamic range of that sample.


3.) 96 < 120


If you want to argue facts #1 or #2, then go find somebody w/ a lab coat to argue with. They came up with the numbers, not me. #3 I don't think anybody could argue.

I'm not sure where you're getting #2 - we were talking about quantization, not aliasing. And technically, 16bit quantization lives just below -96dBFS, averaging roughly -100dBFS.

But you're completely misapplying #3 - unless your peak (0dBFS) level exceeds ~100dB SPL, the quantization error is below 0dB SPL, which is roughly the lowest Sound Pressure Level at which a typical human can perceive sound at the frequencies where our ears are most sensitive in an absolutely quiet environment. 

But in the real world, there is always background noise. And any noise will to some extent mask frequencies at and above the noise frequencies. Note that a noise floor of 20dB SPL might represent an extremely quiet studio environment. A typical home listening environment will be 30-40dB SPL or more.

So you actually need a listening level of well above 100dB SPL (peak, not average) in order to hear 16bit quantization error over the noise floor. If you calculate the peak to average ratio of your audio, as well as the headroom between peaks and 0dBFS, you can find the average listening level this equates to.

In practice, it is certainly possible to hear 16bit quantization error, but you need a quiet room, and either a very high peak to average ratio or a very loud playback level (or both) to hear it.

All of this is well documented and I'd be happy to supply references if you'd like.

pdarg


I am repeating this since I haven't received a yes or no yet:

Oops . . . sorry to start a conflict. But I think that I understand now:

1) "64-bit recording" really means using the Sonar X1 64-bit version in a Windows 7 (64-bit) environment; THAT'S the key to releasing the RAM(?)

2) As far as exporting audio to 64-bit using the 64-bit precision switch within Sonar: the debate rages on; in any case, soundcards can only record at 24 (32?) bit, so in reality, I am proposing to record 24-bit audio using a 64-bit OS/application.

Please correct my misconceptions - or confirm that "by George . . . I think he's got it!"

:)

1. Personally I wouldn't use the phrase "64bit recording", because it's not clear exactly what it means. But I suspect what you state here is pretty much exactly what people would mean by it - recording with a 64bit application on a 64bit OS. And the "64bit" here refers to the the memory addressing, which allows for using tons of RAM.

2. A little clarification here, because there are 2 pieces - Sonar lets you export to a variety of bit depths, including 16bit or 24bit fixed point, and 32bit or 64bit floating point. I can't really see any point to exporting to 64bit files, because a 32bit file takes half as much space and provides all the advantages. So if you exporting something that's a finished product, you use the bit depth you want (16 or 24). If you are exporting something to be imported into another program for further processing, use 32bit.

The "debate" is about whether to use Sonar's 32bit single precision engine or 64bit double precision engine to do the calculations. Theoretically, 64bits means "less calculation errors", but I've argued that in the real world 32bit errors will never, ever, ever, ever, ever, ever, ever be audible, so there's no real reason to use 64bits. Jim argued that there's little cost to using 64bit just in case it might make a difference, so why not use it? I would say it's entirely possible that we are both correct on this one. 

For the second part of your question, yes your sound card outputs no more than 24bits. But since Sonar uses 32 (or 64) bits for its calculations, after mixing and processing tracks you may end up with more than 24bits of meaningful data. My general rule here is to set Sonar's record bit depth to 24bits, since that what your sound card outputs, but set the Render depth to 32bit, since that makes sure you won't lose any data during interim processing.

This has been debated and m*sterbated for years now.  Both sides can "prove" they are right.  Jim Roseberry is the voice of reason here.

vintagevibe


This has been debated and m*sterbated for years now.  Both sides can "prove" they are right.  Jim Roseberry is the voice of reason here.

Yea I know mate it is boring, people who have questions about this should just search through the past forum post on this topic.
 
I am sticking with my 32 bit theory, yea Jim seems to be a knowledgable fellow but I also trust DrewFx as well he seems to have the computing and maths skills to give us fairly accurate answers to these questions.
 
Peace Ben

Muziekschuur at home


To visualize here. You can use the bitmeter here. You can;

Record 24 bit files
Mix with a 64bit engine
Using 64bit plugins at 96khz (they have the two lines)
And write the mix in 64bit (wich will be a large large file)
What will happen is that all processing done to the core audio is done with the utmost precion.
I've used it and you can hear a difference. But it's like the difference between a Dynaudio and a B&W speaker. It's different. But both are good.

When you are gonna do this your source material should all be recorded with the best of the best.
64 bit processing bouncing to a 24 bit file is usually quite good... (enough).

Just do it once. Take a good mix and then bounce to 24 bit and bounce to 64 bit. Put the bitmeter on it and see if the zero's and ones in the 64 bit file move enough.... And do some listening.
In the end that 64bit processing and bouncing can allways be done later. There is no way we can capture audio in anything deeper then 24 bit....

Are you talking about the BitMeter that comes with SONAR, I've never figured out how to use that.
 
Peace Ben