avatar

Cirrus

Tin Man

Posts: 46 Member Since:08/12/2014

#181 [url]

Dec 20 16 10:31 AM

morespaceecho wrote:

Cirrus wrote:
n the case of these amp sim tracks, what if there's 3 of them in the mix?


hahaha 3. there's usually more like 12. or 20:)

Ah, we might be the same person. This happens sometimes.

Quote    Reply   
avatar

tom eaton

Platinum Blonde

Posts: 1,335 Member Since:25/01/2011

#182 [url]

Dec 20 16 12:19 PM

zmix wrote:

Cirrus wrote:Just playing devils advocate here (I love being the dumbest person in the room, right up until my stupitude gets me kicked out of the room...)

When the signal is dithered, there is no truncation distortion. None.  Also dither is by definition, uncorrelated, so adding multiple sources of dither does not add in a linear way.

Exactly,  self-noise can not dither.  Because the self-noise IS the signal.

Quote    Reply   
avatar

morespaceecho

Platinum Blonde

Posts: 2,299 Member Since:29/01/2011

#183 [url]

Dec 20 16 1:19 PM

zmix wrote:
which as we've seen is nearly impossible to properly dither 
 

why do you keep saying this? it's no problem whatsoever. try it.

maybe it would be a problem if you were scaling things up and down 80db repeatedly, i don't know. but nobody does that. 

take a 32fp file. dither it to whatever. line the dithered file up with the original and flip the phase. what do you get? you get the dither noise and nothing else. if there was a problem it would show up in the null.

please try this test yourself before telling me why you think i'm wrong.

www.oldcolonymastering.com

morespaceecho.bandcamp.com

Quote    Reply   
avatar

chrisj

Gold Finger

Posts: 968 Member Since:22/02/2011

#184 [url]

Dec 20 16 3:26 PM

Pretty sure what we mean is: take a 64 bit file (or buss). Dither it to 24 bit, and then copy it to 32 bit (since that's typically done without dither, as it would have to be at all exponent levels simultaneously)

Then null each with the 64 bit file.

The dithered file will null with just noise. 32 bit floating point will have probably a scrunchy sort of truncation noise, quite a lot quieter than that dither noise, but definitely truncation artifacts and not just noise.

If you start with the 32 bit and dither it and null it, you will only hear the difference (only noise). To hear what assigning to 32 bit FP does, you have to start at something higher and null it WITH that something higher, and yeah the truncation artifacts will be lots quieter. Since they scale exactly to the level of the sample, they'll always be less than half as loud as 24 bit fixed truncation and usually quieter than that (only the loudest samples will produce artifacts even close to half as loud as 24 bit fixed)

It's just a matter of degree. You may well prefer it to 24 bit dither noise, as usually it's LOTS quieter, but I have to deal with it every time audio leaves a plugin of mine on AU or traditional VST.

Chris Johnson, airwindows.com

Quote    Reply   
avatar

morespaceecho

Platinum Blonde

Posts: 2,299 Member Since:29/01/2011

#185 [url]

Dec 20 16 6:48 PM

chrisj wrote:
Pretty sure what we mean is: take a 64 bit file (or buss). Dither it to 24 bit, and then copy it to 32 bit (since that's typically done without dither, as it would have to be at all exponent levels simultaneously)

Then null each with the 64 bit file.

The dithered file will null with just noise. 32 bit floating point will have probably a scrunchy sort of truncation noise, quite a lot quieter than that dither noise, but definitely truncation artifacts and not just noise.

 


i did this test the last time we were talking about this, and the 32fp nulled 100% with the 64. 

www.oldcolonymastering.com

morespaceecho.bandcamp.com

Quote    Reply   
avatar

zmix

Aqua Marine

Posts: 4,058 Member Since:20/01/2011

#188 [url]

Dec 20 16 9:14 PM

tom eaton wrote:
zmix wrote:

Cirrus wrote:Just playing devils advocate here (I love being the dumbest person in the room, right up until my stupitude gets me kicked out of the room...)

When the signal is dithered, there is no truncation distortion. None.  Also dither is by definition, uncorrelated, so adding multiple sources of dither does not add in a linear way.

Exactly,  self-noise can not dither.  Because the self-noise IS the signal.

Tom,  we're talking about quiescent noise here, random, brownian, thermal noise...  inharmonic, decorrelated  noise from electron motion...

Quote    Reply   
avatar

tom eaton

Platinum Blonde

Posts: 1,335 Member Since:25/01/2011

#189 [url]

Dec 21 16 1:33 AM

I wish your reference point was a file fading to black rather than a video of a reverb fading into the nosie floor. In any DAW fade out a song while it's playing and see how effectively the self-noise dithers the fade....well, by the time you get close to silence, not only is the self-noise still 80+ dB below the signal, everything moving those last few bits is the music.

Quote    Reply   
avatar

jerry tubb

Gold Finger

Posts: 627 Member Since:06/02/2011

#192 [url]

Dec 21 16 6:02 AM

compasspnt wrote:
I do not worry about these things.

Ha! I don't either really, unless there's a problem to investigate!

making sure the proper Dither button is is activated when bouncing from 24 to 16 in sB

taking care not not to double Dither a DDP in Sonoris

and the occasional null test is about it.

but it's interesting to see just what depths some of us Do think about these things.

i remember reading the textbooks & theories decades ago, and learning to do the right procedures, along w careful listening.

best, JT

Quote    Reply   
avatar

zmix

Aqua Marine

Posts: 4,058 Member Since:20/01/2011

#193 [url]

Dec 21 16 8:23 AM

tom eaton wrote:
I wish your reference point was a file fading to black rather than a video of a reverb fading into the nosie floor. In any DAW fade out a song while it's playing and see how effectively the self-noise dithers the fade....well, by the time you get close to silence, not only is the self-noise still 80+ dB below the signal, everything moving those last few bits is the music.
The example simply showed that the quiescent noise in the recording obviated dither at a truncation as short as 14 bits.  I think that this exmple proves that self dither is possible.

Your argument that applying a digital fade will also attenuate this self dithering quiescent noise is also valid, in theory, but what do you suppose happens to your properly dithered final masters when they're played on a digital audio player, iTunes, VOX , Alexis, iPod, iPhone, etc, and the volume is attenuated by the internal digital volume control?

Quote    Reply   
avatar

maarvold

Aqua Marine

Posts: 3,093 Member Since:23/01/2011

#195 [url]

Dec 24 16 10:25 AM

zmix wrote:

silvertone wrote:
zmix wrote:
morespaceecho wrote:and now for some comic relief, here's a 16 bit mix i was sent today: 

image



32fp would've been helpful:)


The fact that that mix looks like a brick has nothing to do with how many insignificant bits there are in the file, nor does it affect your ability to scale the level of that mix back without any negative consequences...

True Chuck but just looking at this looks like the damage has already been done.  That is the part that novice engineers don't get.


I just finished mastering 300 music cues for Warner Chappell at least 250 of those cues were compressed/limited by a level of 10dB hotter than I've ever delivered a "hot" master.  Just pegging all my meters.  This guy could compose the music but knew nothing about audio recording or engineering.  The director of the project called the files "ugly"... and they were.  So the composer had to give me new files.  Did he take the limiting and compression off and rerun the mixes.  Of course not, he just pulled the level down with all the ugliness intact.  Damage done... next...
Exactly.

I think Scott's implication was that somehow being able to scale the level, as you might in 32 bit float would have saved these mixes.  In reality, and as you've observed, the damage was done long before the output file format was chosen.   If this was done with a brickwall limiter, those usually have provisions to prevent illegal signal levels over 0dBfs, and so there generally  aren't any "overs" in the output.   Reducing the level in your DAW will have exactly the same result regardless of the originating file format.

 
Although the idea of using it on a final mix for release is not a choice I hope I'd be making, iZotope's DeClip (part of RX Advanced) works remarkably well.  The idea is that you decide where you think the threshold of clipping was/is, set the threshold to that value, then apply (negative) makeup gain to make headroom available and DeClip scales the overall volume downward and uses the newly-available headroom to interpolate where it 'thinks' the peaks would have been if they hadn't been clipped off.  It's amazing how, for example, when you match the gains in the unclipped portion of the waveforms, things like big drums actually gain substantial power and guts that was lost due to clipping.  

Quote    Reply   
avatar

gtoledo3

Aqua Marine

Posts: 4,110 Member Since:23/10/2013

#197 [url]

Dec 24 16 11:04 AM

jerry tubb wrote:

compasspnt wrote:
I do not worry about these things.

Ha! I don't either really, unless there's a problem to investigate!

making sure the proper Dither button is is activated when bouncing from 24 to 16 in sB

taking care not not to double Dither a DDP in Sonoris

and the occasional null test is about it.

but it's interesting to see just what depths some of us Do think about these things.

i remember reading the textbooks & theories decades ago, and learning to do the right procedures, along w careful listening.

best, JT

To be super cocky...(hey, why stop now?)

I correctly answered your question pages ago.

Dither when *you* are changing word length. As far as internal ops in a DAW goes, it should already be doing the right thing; applying a non noise shaping dither whenever *it* changes word length for DSP ops.

Converters that capture greater than 24 bit are rare. Converters that play back greater than 24 bit are slightly less rare, but still rare. All that is ever needed to represent 24 bits of dynamic range, is 24 bits.

The point of DSP being raised above 24, is so 24 bits can be yielded on output of a DSP chain. There is *no other reason*. It is *not so that 32 bits can be an end result that people listen to or exchange files of*.

Everything else about it is a bunch of useless chatter.

(Everyone is entitled to an opinion of course, but those opinions can be wrong. The above is STRONGLY my opinon, as well as being factually CORRECT.)

edit: OK, this is overly feisty, but I get fired up talking code...

Last Edited By: gtoledo3 Dec 24 16 11:22 AM. Edited 1 time.

Quote    Reply   
avatar

morespaceecho

Platinum Blonde

Posts: 2,299 Member Since:29/01/2011

#199 [url]

Dec 24 16 11:23 AM

whatintheheckdoyoumean? people exchange 32 bit files all the time (clients sending them to me for mastering for example).

a dsp chain *within the daw* isn't outputting a 24 bit word, it's putting out 32. easily verified with a bitscope. YES I KNOW what is coming out of the converter is 24 bit, no one argues that.

files inside a daw, if you do anything to them, are 32 bits. i simply choose to save them as such, been standard practice here since like 2004. never thought it was any big deal, until these threads.

www.oldcolonymastering.com

morespaceecho.bandcamp.com

Quote    Reply   
avatar

zmix

Aqua Marine

Posts: 4,058 Member Since:20/01/2011

#200 [url]

Dec 24 16 12:05 PM

gtoledo3 wrote:
jerry tubb wrote:

compasspnt wrote:
I do not worry about these things.

Ha! I don't either really, unless there's a problem to investigate!

making sure the proper Dither button is is activated when bouncing from 24 to 16 in sB

taking care not not to double Dither a DDP in Sonoris

and the occasional null test is about it.

but it's interesting to see just what depths some of us Do think about these things.

i remember reading the textbooks & theories decades ago, and learning to do the right procedures, along w careful listening.

best, JT

To be super cocky...(hey, why stop now?)

I correctly answered your question pages ago.

Dither when *you* are changing word length. As far as internal ops in a DAW goes, it should already be doing the right thing; applying a non noise shaping dither whenever *it* changes word length for DSP ops.

Converters that capture greater than 24 bit are rare. Converters that play back greater than 24 bit are slightly less rare, but still rare. All that is ever needed to represent 24 bits of dynamic range, is 24 bits.

The point of DSP being raised above 24, is so 24 bits can be yielded on output of a DSP chain. There is *no other reason*. It is *not so that 32 bits can be an end result that people listen to or exchange files of*.

Everything else about it is a bunch of useless chatter.

(Everyone is entitled to an opinion of course, but those opinions can be wrong. The above is STRONGLY my opinon, as well as being factually CORRECT.)

edit: OK, this is overly feisty, but I get fired up talking code...
Q eff eff E

Quote    Reply   
Add Reply

Quick Reply

bbcode help