Bit Resolution in Audio Processing

The "Merging Cellar" is the place where you can share your tasting experiences and discuss everything from technique, artistic matters or even business practices, but not necessarily about Pyramix. Feel free to pick the brains of the talented Merging forum users. Enjoy.
Forum rules
The "Merging Cellar" is the place where you can share your tasting experiences and discuss everything from technique, artistic matters or even business practices, but not necessarily about Pyramix. Feel free to pick the brains of the talented Merging forum users. Enjoy.

Re: Bit Resolution in Audio Processing

Postby Bernhard Guettler » Tue Jun 03, 2014 16:42

SoundKlang wrote:1) Can you please explain what are the most important parameters in my experiment, and why?
2) 24 bit.
3) Dithering is adding a random signal to the output. Two downmixes with dither would always show at least the difference of the added dither signals. Since we are looking at low level differences here I wanted to listen to the difference in the musical information only. I think that's a good argument for leaving the dither off in this experiment.
4) If we have this difference with levels in the range of only 48 dB, then I wonder what you mean by the term "just nicely" when you said 32 bit float takes care of adding room mics at -110 dB just nicely. Maybe the emphasis was placed more on the "just" than the "nicely"?


In your experiment you are listening to the effects of 24 bit truncation, not to the effects of 32 bit float processing.
Bernhard Guettler
 
Posts: 885
Joined: Thu May 16, 2002 13:35
Location: Berlin, Germany

Re: Bit Resolution in Audio Processing

Postby SoundKlang » Tue Jun 03, 2014 19:16

:lol: Hihi, that was a good one. But seriously, you could read the description of the experiment. Or try it yourself and see if you get similar results.

(One detail was missing in the description from June 2nd and added there.)
SoundKlang
 
Posts: 63
Joined: Wed Sep 21, 2011 18:35

Re: Bit Resolution in Audio Processing

Postby klaukholm » Tue Jun 03, 2014 20:50

klaukholm
 
Posts: 489
Joined: Fri Jul 29, 2005 00:36
Location: Stockholm, Sweden

Re: Bit Resolution in Audio Processing

Postby SoundKlang » Wed Jun 04, 2014 09:11

Now it's not only theory and science but also academic degrees which are more true than real world results?

Let's look at the processing chain: Identical Clip(s) > Gain_1 (ClipGain) > Gain_2 (MixerStripGain, constant) > Gain_3 (MixerStripPanorama, constant) > Gain_4 (MixBusGain)

We have an identical sound source. This is processed through exactly the same functions which are not adding any kind of randomness. With this identical chain we make two versions which only differ in the parameters for Gain_1 (ClipGain) and Gain_4 (MixBusGain) :

1. version: Gain_1 = +12 dB , Gain_4 = -12 dB
2. version: Gain_1 = -36 dB , Gain_4 = +36 dB

In both cases we have the same overall gain. At least theoretically both versions should be identical.

If now we cut off the lowest value bits of both versions, keep only 24 higher value bits, and both versions still show a difference, how are we then listening to the result of truncation?

If you guys would really want to know, you would have tried this yourself with different target formats. And if not my Pyramix version is different from any other, you would always get the result that both versions differ. But you don't want to know.

What you think, shall we stop wasting our precious time on such fruitless debate?
SoundKlang
 
Posts: 63
Joined: Wed Sep 21, 2011 18:35

Re: Bit Resolution in Audio Processing

Postby Graemme » Wed Jun 04, 2014 09:39

SoundKlang,

Lay off the sarcastic comments...getting tiresome and you seem to miss the point that people are actually engaging with you...
Graemme Brown
Zen Mastering
1460 Wild Rose Drive
Gabriola Island, BC
Canada V0R 1X5
+1.604.874.9096

"A Horus, A Horus; My Kingdom for a Horus!"
User avatar
Graemme
 
Posts: 2195
Joined: Wed May 15, 2002 22:18
Location: Vancouver, BC, Canada

Re: Bit Resolution in Audio Processing

Postby SoundKlang » Wed Jun 04, 2014 16:22

Very well then in plain language: Nobody invited them to do so.

And just to make it more clear - the last thing I need is some zen or scientific mindset turning over ten years of careful preparation, hard work and an insane amount of patience and dedication into garbage.
SoundKlang
 
Posts: 63
Joined: Wed Sep 21, 2011 18:35

Re: Bit Resolution in Audio Processing

Postby Bernhard Guettler » Wed Jun 04, 2014 23:46

Soundklang, you sound terribly angry at science. That's a bit insane, considering that without it you would not be talking about this subject here.
Why don't you just run the experiment yourself again mixing down to 32 bit instead of 24 bit, eliminating your faulty premise that makes audible the 24 bit truncation artifacts first of all?
We will tell you what will happen: You will still have a very small audible difference, but it will be much smaller and with less distortion.
You will not surprise anyone with the realization that there are very small computational errors in the 32 bit float math. All the big boys knew that already. (I'm not one of them, DSP is not my specialization.)

What you will have to show in order to convince anyone that the implementation of a higher bit depth for internal DSP is so urgently necessary in general and not only for certain special DSP tasks, is that the verifiable difference is a perceptually relevant one, also for simple gain calculations. If you are very good you could be successful. So far you have shown nothing of interest.
Bernhard Guettler
 
Posts: 885
Joined: Thu May 16, 2002 13:35
Location: Berlin, Germany

Re: Bit Resolution in Audio Processing

Postby SoundKlang » Thu Jun 05, 2014 09:22

I am not angry at science. But I find it surprising when people in general, including scientists, are so convinced they know already everything they need to know, that they start to argue in form of unproven statements (noise) instead of looking at the question (signal) - maybe a distraction strategy?

So far you have shown nothing of interest.


So true for you, at least regarding this "dialogue".

DSP is not my specialization


Probably, because either what you mean by "truncation" is different from what on a technical level it should be, or it seems you are not able to think at the bit-level. It may well be, when analyzing the audio of a truncated signal, one can find some difference (frequency, phase, distortion or whatever) compared to the original signal and then this difference is called truncation artefact.

But here we are looking at bit operations. Throwing away the lowest bits does not alter the higher bits in any way. So if that is done to two identical signals, the higher bits of both signals are still identical. If then after the truncation the identical higher bits of both signals are subtracted (phase inverted summation) they null out. Total silence, Null. You can amplify (=multiply) Null not only by 96dB but whatever you want and still get Null (=total silence).

Which means: The error we are listening to is not a result of truncation (if that word still means what it should on a technical level). I can not say if the error is caused by 32-bit floating point processing in principle or if the cause is some implementation detail in the signal flow of Pyramix, maybe a difference in the implementations of ClipGain vs. Fader-Gain in the mixer.

just run the experiment yourself again mixing down to 32 bit instead of 24 bit


There must be some reason, why you continue to not read carefully.

... small computational errors in the 32 bit float math ... will have to show ... that the verifiable difference is a perceptually relevant one ...


In my mixes the difference comparing the different versions of the mixdown was not small, it was far more obvious than the difference between dither on/off. The question is, what is "relevant"? 24-bit audio playback was not "relevant" when the CD was introduced. Some audio professionals do not use dither, because for their results it is not relevant. Some audio professionals do employ digital clipping (to a certain degree), some purposely because they like it's sound. Some people use freeware audio sequencers, because for them leading edge sound quality is not relevant.

I do not have to show anything. I am not doing the work of other people here for nothing. This was a feature request. I tried to give some arguments for why this feature could be interesting. Some people find ridiculing this feature request more relevant than the request itself.
SoundKlang
 
Posts: 63
Joined: Wed Sep 21, 2011 18:35

Re: Bit Resolution in Audio Processing

Postby Bernhard Guettler » Sun Jun 08, 2014 10:17

Do the test with mix down to 32 bit instead of 24 bit and maybe you will begin to understand a bit. But probably not.
Bernhard Guettler
 
Posts: 885
Joined: Thu May 16, 2002 13:35
Location: Berlin, Germany

Previous

Return to The Merging Cellar

Who is online

Users browsing this forum: No registered users and 2 guests

cron