Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Timmy]

Take a 24bit 44.1kHz wave file and name it A. (please be sure that it is a real 24 bit file but not a fake one that converted from a 16 bit file) and you’d better use a stereo one because some workstations have different pan laws.
Ok now let’s go.

Use your workstation to convert (u can Bounce, Mixdown, Export or something else) file A to 16bit and we name it B.
Then convert A to another 16bit with Dither and we name it C
Maybe once again with another kind of Dither (Pow-r or Apogee UV22 or WAVES IDR) and get a file D.

Well now take B C and D, convert it back to 24 bit and we get B’ C’ and D’

Pick Any Two from A to D’ into your workstation and do a phase reverse on any one of the two and listen to the output. What it might be?

I did this on Cubase SX and Samplitude and it’s always Silence. Isn’t there any loss during the conversion?

Unbelievable? Just try it out yourselves.

[ammitsboel]

There is a loss.
Did you pull up your fader? did you try to normalize your output?

[Timmy]

Have you try it?

After dithering and bit changing, all files are still same!!!

Phase reverse one and it will cancel another.

[stevieeastend]

Hi,

The goal of a correct dithering and SRC is to have the same material just in a reduced bit length and it depends on the different digital processes which sound results you will get.

As far as I understand the whole thing is that there can not be a difference because the material stays still the same, the files are just put together/reordered differenty. It HAS to be the same in terms of a correct dithering and SRC. If not than the algorithm sucks.
That they all SOUND different is because the different dither algorithms got their own concept of how the 24bit material is put together/reordered in 16bit.

Let me tell you about a similar experience:  I exported 32bit floating files out of Sequoia as 24bit files, one as AIFF, another as .WAV in order to send them to mastering.
Then I imported them back into Sequoia, put them under the original file, phase reversed them: result was silence. They look like all the same. They are the SAME, the difference in sound is due to the different ways how the material is computed to get the same result...

This is my personal approach as a musician, producer, arranger. If you wanna have a correct technical explanation I am sure Nika can give an answer you surely will never forget

cheers
steveeastend

[maxdimario]

Aha!
the dreaded phase reversal test!

well it seems that digital processes are famous for not distorting the sound, and the proof is in this phase reversal test in a daw! I've seen many posts that boast the superior quality of digital mixing by the above test.

the mistery of digital!

I remember deciding whether to bounce a vocal comp or not (in 16 bits to 16 bits) because we noticed that everytime something was bounced it became duller and fuzzy-er.  Something does happen to the sound.

could it be that 24 bit workstations are taking liberties with their bit-depth processing for playback, without telling us anything?

this would explain why out of phase signals transferred to 16 bits and back cancel out completely when summed out of phase with the original 24 bit.

[moadriaanse]

I have also conducted this test in Pro Tools, but with an 88.2/24bit session, bounced to 44.1/16 stereo, and imported back into the 88.2/24 session for "the null test".
The only difference was a littlebit of (dithering) noise, and - slightly above the noise - an occasional bass thump. To me the difference wasn't significant; I've been very happy with "Bounce to Disk" since I did that test.
Best Regards,

Martin

[David Schober]

The theory of the null test is simple enough.  However I've wondered if many of these tests are truly legit.  I haven't done the leg work for this, but it wouldn't surprise me a bit to find fairly audible differences can exist despite the fact that the files apparently cancel each other out.  If the only differences were in extreme high frequencies, I'm not sure one could discern them.  Amplifier and system noise alone would probably mask it.  If one were to take an audio file of a mix, then lay down the exact same audio file, but this time added a dB or two up at the top for a little more "air" and do the null test, I'd be surprised if that difference would be audible.  In a very, very quiet system with an top of the line D/A in a quite room with a speaker system capable of revealing such thing it might be possible.  It's more likely that the differences would be seen in a high quality digital meter, but you'd never hear it.

The problem with doing tests like these on the kinds of gear home studios typically have is most systems simply can't reveal those very small, but possibly very important differences.  The resolution just isn't good enough to show what's going on.  

[Timmy]

Thank you all guys!

I used a very accurate meter and finally saw the error sighal between the Dithered files. It was at about –90db! Isn’t it amazing? –90db is much lower than the self noise of most monitoring systems.
It do change my conception of Dither.

Ok now leave the Dithered files apart, let’s check the ones without dither.
And something even more amazing comes.
When I compared my file A B and B’ , the very accurate meter read a -∞.  
You know what?That means after transforming a 24bit file to a 16bit file and then transform it back to 24bit, even an accurate digital meter can’t tell the difference between these two 24bit file!
Some one suggested me to use a bitscope, but I think it’s enough. We don’t want to See a difference, we want to hear. Don’t we?

Well, A 24bit file is about 1/3 times larger than a 16bit one, what a wast.
I think from now on we can record at 16bit192KHz?  

[Bobro]

Timmy wrote on Sat, 12 February 2005 11:24

Have you try it? After dithering and bit changing, all files are still same!!! Phase reverse one and it will cancel another.

No it won't. Normalize your file and take a listen, as Henrik said, above your post.

Listen to the nature of the noise, the color of it, as well. I do quite a bit of work with noise, doing synthesis- it's very instructive to listen to different kinds of noises (not too loud of course!).  Dithering noise, self-noise of equipment, the noise generator of a synth, wind and rain... there's a whole world of different kinds of "noise".

Certain synthetic noises are very natural and sweet, you can solo on them with nothing but a resonant low pass filter to great effect. Mic a rainstorm with a stereo pair and put it in your recording- it's amazing how loud it can be before anyone even notices it, or is bothered when they do.

Other noises are just nasty and insidious.

And when you add up track after track of any particular noise?

Well those are my opinions.

-Bobro

[ammitsboel]

Timmy wrote on Sat, 12 February 2005 11:24

After dithering and bit changing, all files are still same!!!

No they are not!
listen and you will find

[maxdimario]

..or it's the workstations that are doing something funny.

a 24 bit signal with no dither converted to 16 and back to 24 is not the same signal unless:

a) the 24 bit signal is not a 24 bit signal, but only has the format of 24 bits with 16 bit information.

b) the summing bus or signal 'path' of the daw is not really full 24 bit resolution.

c) some other obvious explanation that will make me feel that digital audio is a predictable and secure medium.

I can take an analog mixer apart to see exactly what's going on but I can't take a program apart and analize if it really does what it says it does on the brochures.

[Paul Frindle]

Timmy wrote on Sun, 13 February 2005 18:55

Thank you all guys! Ok now leave the Dithered files apart, let’s check the ones without dither. And something even more amazing comes. When I compared my file A B and B’ , the very accurate meter read a -∞.   You know what?That means after transforming a 24bit file to a 16bit file and then transform it back to 24bit, even an accurate digital meter can’t tell the difference between these two 24bit file! Some one suggested me to use a bitscope, but I think it’s enough. We don’t want to See a difference, we want to hear. Don’t we? Well, A 24bit file is about 1/3 times larger than a 16bit one, what a wast. I think from now on we can record at 16bit192KHz?

The 24bit file must have already been a 16bit effective source (i.e. 16bit mapped onto 24bit format) or has been truncated within the W/S to 16bits internally before converting. If you haven't changed anything within the W/S like gain or EQ, no extra information has been added that might occupy the silent range below 16bits.

Ok now let me go over what you have done (correct me if I'm wrong please)

a) You got a file which was originally (supposedly) 24bits.
b) Converted it to 16bits without dither - thereby losing the bottom 8 bits - if they were active.
c) You converted the original file again to 16bits - thereby losing the bottom 8 bits - and then reconverted to 24bits once more. Therefore only the top 16bits of the new 24bit file contain data and the rest is present but null.

You take away a)or b) from c) and get absolutely nothing at all. The difference ought to be distortion and noise around -90 - 93dB - but you have complete silence.

This means that ONLY the top 16bits were active in the original file or by the time the W/S saw it (or within the W/S itself) before conversion. There is no other explanation. The bottom 8 bits could have been lost at some stage throughout the acquisition process (sound card settings perhaps or an error therein) - or may have been missing in the first place.

My experience of some of the applications you mention (particularly those running on PCs), is that it is very difficult to get a signal through them and the hardware without any digital error at all. There are far too many variables and foibles in the platforms, 3rd party interfaces and host S/W apps. Sadly IME the only popular application I use regularly that achieves this routinely correctly is ProTools using their own interfaces and H/W.

[tamasdragon]

I you have good quality equipment from the first place to the last, you operate it as you should, than I think it's very hard to screw up something, whether it's pc or mac or whatever.
Regards Tamas Dragon

[Paul Frindle]

Fenris wrote on Fri, 18 February 2005 00:59

Quote: My experience of some of the applications you mention (particularly those running on PCs), is that it is very difficult to get a signal through them and the hardware without any digital error at all. Can you elaborate? In my experience with Nuendo 1 and 2 and Cubase SX on Windows XP, on a properly outfitted system, I've NEVER experienced corrupted audio. No clicks, no dropouts, no dropped bits. Wavelab has an Analyzer tool which tells you the "apparent resolution" of the file, ignoring empty bits, and my 16, 24, and 32-bit files are always what they should be. Cubase VST with first-generation Lightpipe interfaces was a different matter. But that was five years ago. This is pure hearsay, but I've heard that Pro Tools LE isn't fully 32-bit like other native apps.

It has to be said that some of the systems I have used are indeed up to 5 years old. But the condition I'm talking about is with an Audio Precision test set connected digitally input and output. I need to do this to test plug-in S/W since there are no tools up to the job within the S/W platforms themselves. Any dropped, repeated or corrupted sample ruins the test. Running say and FFT that needs to operate absolutely correctly during a 10sec period is a real challenge IME - an error of some kind will occur within this period around 20% of the time. Now this may not seem much - you certainly might not hear it - but it's indicative of instability and problems none the less.

I don't imagine for a moment that this is the fault of Nuendo, Logic or any particular host S/W - that would be very unlikely, since messing such things up would be difficult and indeed costly. It seems IME to be a problem with the accumulated issues associated with windows (which isn't a true real time operating system), ASIO drivers, audio interface systems and the associated drivers, synchronisation and interactions between the various processes that are running.

It is clear from the this thread that someone has recorded what they thought was a 24bit file that somehow has ended up with only 16bit data in it. This sounds to me like just such an issue that could occur when interfaces between different parts of the system are running at odds with one another. For instance the sound interface card running a 16bit format but the host S/W reporting 24bit operation. I have had this happen (and even worse) in my battles to get digital conformity through some systems It would seem very unlikely that I was the only one to suffer such experiences and it seems entirely possible that people have indeed made projects with issues like this present - without actually noticing.

Doing such testing with PT, you connect and it just works - indicating the presence of a stable system and operating environment. I could certainly become confident enough to use such a system without feeling an urge to test it each time to confirm it was actually operating correctly

[tanov]

I think that the goal of the 24bits processing is in
editing (not the final mix).

Here is an example:
1) You have to make some calculations with calculator set to
round numbers to the first digit after the floating point.
Write the result.
2) Then disable the round function.Make some calculations and
finally round the result. Now compare it with above result.
You'll find the big error in 1)

The same case is in a DAW. The DAW rounds calculated bits
up and down.But 24bits rounding is smaller.