The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Good dither practices, what are yours? Dynamics Plugins
Old 9th March 2017
  #361
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by FabienTDR View Post
Given background noise in the signal above dithering level (very likely the case with music signals, even down to 16 bit), dithering is as wise as insisting on full moon before rendering.
Nope. As I've said before, I can tell 24 bit noise from 2 bit noise 10 out of 10 all day long… IF it is not dithered. If you TPDF dither it, it becomes impossible to tell.

'self-dither' is wrong, and you're wrong on this subject. I did this experiment using ABX software and felt quite pleased with my ability to infallibly tell when it was low-bit. Dither, correct dither, immediately changed that. You're wrong here.
Old 10th March 2017
  #362
Motown legend
 
Bob Olhsson's Avatar
 

According to Jim Johnston, if you fail to dither, the math is simply wrong. He's a fellow of both the IEEE and the AES based on decades of work with digital audio at Bell Labs where digital audio was invented.
Old 10th March 2017
  #363
Fine, let's try! I built a reasonable test file (took me hours). But feel free to suggest your own.

Just load the file and press render, then post it here.

Effects will be clearly visible, but will they in the case of the noisy section or when exporting to 24bit?
Old 10th March 2017
  #364
Bob, with a signal having a noise floor far beyond the minimum dithering level, you are dithering whether you add a minuscule amount of noise or not.

I'm not arguing against dithering. I'm just saying that in 99% of the cases, you have already. The dithering step then becomes a pure fantasy bias fest. Especially when working with music targeting standard audio rates.

Truncated noise is noise. It doesn't change its properties during the process, and is the central reason why dithering works in the first place. The whole *magic* behind dithering is the fact that noise is immune to truncation distortion. You can beat, insult and squeeze noise like crazy, cut a leg, an arm away, it's still noise (Monty Python, yes ).

That's why, if ever needed, you make sure to only truncate noise (i.e. random content) and keep stationary signals at a safe distance (which in turn are very sensitive to any form of nonlinearity: they immediately create new and potentially audible partials).

Last edited by FabienTDR; 10th March 2017 at 01:46 AM..
Old 10th March 2017
  #365
I'm not trying to start a rebellion. In fact, all this stuff was the result of an attempt to work on a cool dither for modern use. My conclusion was a bit depressing. I really wanted. Given modern workflows and half way typical music material, "No dither" is by far the cleanest option. Truest to the source.

Before, I never thought about the nature of music signals vs dithering considerations and simply accepted the TPDF default case as a given best practice. But the most ironical thing is, the more you leave the isolated lab cases of truncating a sine to 8bit or 12bit, the more it becomes surprisingly irrelevant. Almost humorously irrelevant

Now thinking about a tool warning the operator when the signal isn't sufficiently random! Maybe that helps reducing doubts. From what I'd expect, 24bit, 20bit and 16bit will flash green most of the time!

Last edited by FabienTDR; 10th March 2017 at 01:49 AM..
Old 10th March 2017
  #366
Motown legend
 
Bob Olhsson's Avatar
 

If no dither sounded better, something was broken. I often can't hear 24 bit sounding better but I've never heard it sound worse unless there was a problem.
Old 10th March 2017
  #367
Lives for gear
 
JP__'s Avatar
 

I think it has also to do with socialisation; some prefer the sound of truncation (or aliasing etc). I had to learn that kind of "clean" sound for myself, to be true.
Old 10th March 2017
  #368
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by FabienTDR View Post
Truncated noise is noise. It doesn't change its properties during the process, and is the central reason why dithering works in the first place. The whole *magic* behind dithering is the fact that noise is immune to truncation distortion. You can beat, insult and squeeze noise like crazy, cut a leg, an arm away, it's still noise (Monty Python, yes ).
No, it's not. As I said, I ABX tested that. I don't know where you're getting this but it's not double-blind testing. The most startling thing is, this IS the only true purpose of double blind testing.

If folks are going to invoke double blind testing to try and prove the absence of a thing (hearing subtle differences, which double blind testing does NOT prove) but refuse to accept it when double blind testing does 10/10 prove the existence OF a thing (truncate noise hard enough and you change it audibly), that's not science.

And again, to have this be obvious you'll want to use a very low bit depth rather than trying to hear something like 24 bit vs. 8 bit, and you cannot TPDF dither it or you won't be able to hear a difference anymore. I learned that, and I was strongly motivated to also pass the dithered tests, but there was no chance of that. I could only hear an obvious difference on purely truncated noise.

And, that's not even how dither works. Dither statistically distributes quantized values across a range of output values to approximate the input value when averaged. Noise ain't got nothin' to do with it, you can pursue different methods of doing that thing, and I have done. The key factors are producing this distribution of values, and keeping the artifacts (such as added noise) consistent and free of distracting behaviors. Using two noise sources 1 LSB in amplitude happens to do both of those things, but if you get the amplitude wrong it's not dither.

Dither does not work by raising the noise level above the point of truncation. Dither works by MATCHING the range across which truncation happens (flat dither) causing this statistical distribution of values, and then by decorrelating the resulting noise amplitude (TPDF dither) from fluctuations that relate to the input signal.
Old 10th March 2017
  #369
Lives for gear
 
stinkyfingers's Avatar
 

Quote:
Originally Posted by FabienTDR View Post
Truncated noise is noise.
...coming from someone with a logo/name inspired by japanese culture...
Old 10th March 2017
  #370
Who said not to dither?

I clearly said: You are already dithering, if you want it or not. In no case I was debating dither vs no dither. Please read more carefully.

I wouldn't have thought that something as trivial and simple as dithering to become such a personal thing for many. Now don't drag me into this childish personal area, the affection to a random number generator is a personal "thing" (lol). I'll leave you alone with it, no worries.

Last edited by FabienTDR; 10th March 2017 at 11:38 AM..
Old 10th March 2017
  #371
Quote:
Originally Posted by stinkyfingers View Post
...coming from someone with a logo/name inspired by japanese culture...
Sorry, I don't get this. I was talking about the fundamental mathematical properties of noise. They are the only reason why dithering works in the first place, you can beat it, you can wipe it out or cut it in two. It's still noise with exactly the same properties, information and sound. This is the most central "thing" behind dithering, the only reason why it works. But obviously, most prefer bureaucracy over understanding (engineering). It's so much easier!

I'll better pack this into a blog post with all proofs. Here's obviously a harsh place for any thoughts going beyond the "norm".

Nobody felt the need to support his claims via the test file (or any). I find this quite revealing! (in particular from engineers specialized on processing stereo files).

Dither noise is nothing magical. We specify it this way (at least +/- 1bit) because we assume that less noise is simply better (in the THD sense). But that doesn't exclude that a 0dBFS noise signal wouldn't be an equally capable dither. The only relevant point is the randomness. The "removal", the stuff we throw away, has to be fully random to prevent loss (in the harmonic structure). That's all. If you don't get this, you don't understand dithering.

Chrisj, I wonder about your own understanding of the process, now that you told me that noise doesn't make a dither. It's as wrong as it can be, and you seem so certain. Again, watch what you remove during the process and you'll understand that it is purely about the randomness of the removed data. If it isn't random, partials appear, the harmonic content changes(= irreversible loss). With a given noise floor above dithering level, again, a very likely case, dithering is implicit.

The level of the noise doesn't matter as long it's AT LEAST +/- 1bit. That's rather easy to prove, just try to truncate noise and show me the distortion... Bam, end of discussion. ...Why? There is none of course, as the signal is sufficiently random over the truncation area. Dithering was successful, we have a clean, transparent truncation. And this without adding dither plugin XY.

Last edited by FabienTDR; 10th March 2017 at 01:43 PM..
Old 10th March 2017
  #372
Lives for gear
 
JP__'s Avatar
 

Quote:
Originally Posted by FabienTDR View Post
Nobody felt the need to support his claims via the test file (or any). I find this quite revealing! (in particular from engineers specialized on processing stereo files).
!
The only thing thats revealing is your ongoing arrogance. I have done my tests ( as well in coop with some signal theorists). And I personally claimed it before: it all hugely depends on several factors how audible the truncation distortion is in a real life scenario. So personally Im just not really interested to join any "official" tests as the results will mean nothing for my work. And Im also not really interested to share my results with the public (as I used clients music for this).

I see no harm in a tiny bit of added noise (even when useless), but I see a lot of harm in unwanted distortion (even when its not audible at first). So all those theoretical arguments regarding the use of dither simply means nothing for practical thinking ppl.
Real life test scenarios only will work when we include the whole processing chain. Just to pick out a single part of that chain will not show any reasonable (in the sense of real life application) results anyhow. Thats the mistake all theoretics here do constantly over and over again...
Old 10th March 2017
  #373
Then show your tests. What's so hard about it?! I'll bet my lunch that you won't anyway.

About "signal theorists". Again, the irony with dither theory is that it only works in isolated lab cases: Truncating a sine to 8bit or truncating a sine to 12 or 16bit. With real world signals having typical noise floors and complexity, adding a minuscule amount of noise won't prevent anything. At 24bit anyway. Do you get the irony? I'm asking for practical proofs, you guys obviously fail at it and defend yourself calling me funny things like a "theorist".

I'm also the only one asking for practical tests, not the snake oilish rhetorical trickery I'm messing with over the last posts. But it's me the signal theorist? Guys, seriously. I even provided a test signal, and I'm a theorist? That's what makes me so angry, the personal attacks free of any basis. Why so personal?!

I'll now politely leave the place if you allow, and better post my "blasphemy" elsewhere to not hurt feelings. Happy THD reduction!

Edit: For completeness, here the original file and test cases to 16bit and to 24bit. Truncated without explicit dithering. Open a FFT analyzer, look at the "non theorist" segments (2, 3 and 4) using a wide window size, compare and draw your own conclusions.

http://www.tokyodawn.net/labs/public...tSignal_v1.wav
http://www.tokyodawn.net/labs/public...ncatedTo16.wav
http://www.tokyodawn.net/labs/public...ncatedTo24.wav

To 16bit, I see an irrelevant amount of quantization distortion due to the lower noise floor at HF (this is brown noise, like typical electronics produce. Its level is lower at HF, lower than the 16bit minimum noise level, hence the slight quantization). With a white noise bed, you'd see nothing at all.

To 24bit, I see nothing. Not just low distortion, I see no distortion at all (this one doesn't accumulate in later stages, as it doesn't exist. It's fully dithered).

Further, the more natural the signal, the less it exhibits quantization distortion (last segments in the test file become increasingly "natural"). It is clearly visible and measurable. This directly supports what I said above, despite you guys' "no" "nope" "you don't get it" "I hear" "it's wrong". Hold on, maybe you learn something today.

we've left theory! Now it's your turn. Please no more personal, unverifiable anecdotes.

Last edited by FabienTDR; 10th March 2017 at 03:13 PM..
Old 10th March 2017
  #374
Airwindows
 
chrisj's Avatar
"But that doesn't exclude that a 0dBFS noise signal wouldn't be an equally capable dither."

It's not. This is the one thing double blind testing is actually good for. I can truncate noise and tell it blind in ABX testing. I've done so. Then, challenged to do the same when the noise is dithered, I confidently expected to do exactly the same and completely failed, which rules out psychic powers and magical guessing luck.

I can do a lot better than TPDF dither and have done so. This is my most recent work in the field, a Patreon-supported free AU/VST called NotJustAnotherDither by popular demand (I ran a contest for naming it). I use professional mastering engineers for feedback and correction, which has been hugely useful: I wouldn't have NJAD if Bob Ohlsson hadn't goaded me into revising 'Naturalize' by finding some legit problems worth renewed attention.


I promise the mathematical and functional underpinnings of dither is NOT 'the background noise is louder than the truncation threshold'. I'm certain because my reasoning on the common basics of digital theory is backed up by a lot of qualified people up to and including JJ (whom I think Bob O. mentioned), and is not really different from their consensus (it's your take on it which is more unusual, and you're not aware of this). When I do step beyond the understood basics of digital theory, such as with the usage in NJAD of noise-shaped Benford Realness calculations, it's with an awareness that I'm doing something unusual that's not been done before and the understanding that further experiment might reveal weaknesses in my methods, as in fact Bob O. did, to my eventual benefit.

This:
Quote:
Again, watch what you remove during the process and you'll understand that it is purely about the randomness of the removed data. If it isn't random, partials appear, the harmonic content changes(= irreversible loss). With a given noise floor above dithering level, again, a very likely case, dithering is implicit.
The level of the noise doesn't matter as long it's AT LEAST +/- 1bit. That's rather easy to prove, just try to truncate noise and show me the distortion... Bam, end of discussion. ...Why? There is none of course, as the signal is sufficiently random over the truncation area.
This is mistaken. It is not 'end of discussion' because I've passed that 'try to truncate noise and show a difference' in double blind testing for YEARS now. Does it constitute harmonic partials? No, it's still noise, but when noise is a higher amplitude it is the signal you are trying to preserve, and it has to be dithered (with other, specific noise, OR with some other wordlength reducer such as my NJAD) or it will sound audibly worse when truncated.

I would like to hear your work, though given that you're finding truncation preferable, maybe that's a silly thing to ask. I need a lot better than truncation, and I'm not the least bit concerned with 'modern mix styles': to me, that means bucketloads of poorly-coded and often rented plugins producing a sound already so degraded in just the areas truncation damages (through extensive repeated truncation at every stage of internal processing) that it's easy to see why you don't feel you're gaining anything from a final TPDF dither. Also, by 'modern' you're clearly not thinking of artists like DeadMau5 who are working with analog synths and extremely high quality conversion and processing at every stage, whose output would be severely degraded by digital truncation… and who perform at volume levels where this would be easily perceptible. You'd better be dithering your output if you're playing as loud as DeadMau5 is, on PA systems as good as he uses!

And the presence of continuous repeated truncation when returning to 32 bit floating point for the CoreAudio or VST or Pro Tools buss, is something we've talked about extensively in this thread and is another fact relating to the nature of digital number storage, but if you're not getting how truncation to fixed point works, it's going to be even tougher to persuade you that floating-point truncation is a thing. (I'm really, really gonna need to make that video explaining this topic)
Old 10th March 2017
  #375
Only anecdotes, no logical explanations, nothing. As expected.

Still waiting for your truncated noise example. It's so obvious, eh? Then do it now or accept being called a misslead theorist!
Old 10th March 2017
  #376
Lives for gear
 
JP__'s Avatar
 

Come on Fabien, please do not continue to make those stupid claims. You are much smarter than that, I know that.
I made my points clear, please accept that as I accept your points as well.
You ask me for my test and I said no (with good reason as mentioned above), please accept this too.
You are good in making your own tests, we all know. And if you come to other results than me, I can accept that too.
Im not here for proves, Im here just for fun.
In the end of the day I tune my masters by ear; if that means I use gear xyz or this EQ setting or I just engage a dither plugin, because it sounds better to me. If the setting is different for another master the next day Im absolutley fine with it.
These constant proves whats wrong or right feels like religion to me, this counts for all partys here.

To say it once again: Im pro dither because the noise of my analogue chains covers up all dither noise at all circumstances, but the analogue noise does not cover up any truncation distortion at all circumstances in my experiences. What else do you want me to prove here?
Intetestingly this was more obvious with my old converter setup. And this is what I mean withn many influences that plays a role here.

Last edited by JP__; 10th March 2017 at 03:15 PM..
Old 10th March 2017
  #377
I don't accept the no. Nothing's easier for you than exporting audio to n bit. I even did it for you, see above. This is audio engineering, not about writing poems. Now you prefer sticking to theory? Is that cool? I'm also here for fun, but that isn't.

I accept the idea that this noise doesn't hurt anyway. But don't insult those who are not happy with it and come up with perfectly logical and verifiable explanations allowing to avoid this often confusing step.

Dithering can be explicit (as "ordered") or implicit (due to the natural noise content of the file). Both is perfectly valid dithering, my tests above (and theory) clearly prove it.

Last edited by FabienTDR; 10th March 2017 at 03:54 PM..
Old 10th March 2017
  #378
Lives for gear
 
Plush's Avatar
Bickering over a minus 160 dB noise?

Perhaps it will always continue.

Noises out of Germany in particular.

A fetish for exactness.
Old 10th March 2017
  #379
qwe
Lives for gear
 

Quote:
Originally Posted by chrisj View Post
I promise the mathematical and functional underpinnings of dither is NOT 'the background noise is louder than the truncation threshold'.
Perhaps the following is of relevance...

From this 1992 Lipsh*tz paper (Gearslutz strips out the sh*tz in the link as well. :-()

"We recommend the use of nonsubtractive, triangular pdf dither of 2-LSB peak-to-peak amplitude for most audio applications requiring multibit quantization or requantization operations, since this type of dither renders the first and second moments of the error signal constant with respect to the input, while incurring the minimum increase in error variance."

From the book "Applications of Digital Signal Processing to Audio and Acoustics" (Edited by Mark Kahrs and Karlheinz Brandenburg.)

Google Books preview: https://books.google.co.uk/books?id=UFwKBwAAQBAJ

From p199:

"Dither. Starting from Robert's pioneering paper [Roberts, 1976], the use of dither in audio was seriously analyzed by Vanderkooy and Lipsh*tz [Vanderkooy and Lipsh*tz, 1984]. The basic idea is simple: to whiten the quantization error, a "random" error signal is introduced. While the introduction of noise will make the signal "noisier," it will also decorrelate the quantization error from the input signal (but not totally). Vanderkooy and Lipsh*tz also propose the use of trangular dither derived from the sum of two uniform random sources [Vanderkooy and Lipsh*tz, 1989]."

Also: https://www.gearslutz.com/board/5985683-post144.html
Old 10th March 2017
  #380
Lives for gear
 
JP__'s Avatar
 

Fabien, Im not any interested in your test, if Im curious I make my own in my environment within my chain.
I have added some points to my post above. And I think I made reasonable clear why I cant share my testfiles. And Im not interested in doing new tests as Im through this topic for now.

In my experiences it was clearly shown that the self noise of the analogue chain will not act as self-dither as other have mentioned correct that you have to dither BEFORE truncation. So the analogue noise acts like noise only. Sound logical to me, at least after my tests.
Old 10th March 2017
  #381
Quote:
Originally Posted by JP__ View Post
In my experiences it was clearly shown that the self noise of the analogue chain will not act as self-dither as other have mentioned correct that you have to dither BEFORE truncation. So the analogue noise acts like noise only. Sound logical to me, at least after my tests.
But I demonstrated the opposite above. Just download the files and look at them.

The 2LSB or +/- 1LSB is a minimum value defined with THD in mind. It's optimal in the sense that it doesn't raise the noise floor beyond what's really needed. More noise doesn't suddenly break the randomization process, quite the opposite. As soon you move beyond these minimum limits (-60dB is far beyond 2LSB, isn't it?), the nature of the noise doesn't matter (TPDF, gaussian, or whatever), truncation is clean, no measurable partials appear. You can verify for yourself.

Even the 7 sines without noise do something interesting: no measurable partials appear, if anything, the truncation raises the noise floor! (in some sense, it's self dithering due to signal complexity). This makes me wonder which real world audio file really needs explicit dithering (to 24bit or 16bit), it looks more like an exception to me, not the standard case. Music signals consist of far more than 7 frequencies.

Last edited by FabienTDR; 10th March 2017 at 03:54 PM..
Old 10th March 2017
  #382
Quote:
Originally Posted by Plush View Post
Noises out of Germany in particular.
???
Old 10th March 2017
  #383
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
truncation is clean, no measurable partials appear.
Harmonic error is only one form of quantization error?
Old 10th March 2017
  #384
New partials (harmonic or not) only. Be aware that they can sit right on top of another original partial, but this is unlikely with this type of non harmonic test signal.

Without new partials appearing, there can't be nonlinearity in the system. The original information is intact.



As mentioned several posts above, a nice trick is to imagine what happens to the signal that we remove. If it contains any regularity (= a stationary signal), new partials will appear due to the cuts. If it is purely random (doesn't contain any regularity), there's no partial appearing. Again, beaten noise is noise (not a joke, it's very relevant). That's the magic, it's still as random as before, this property can't be changed by any process, noise is literally immune to any form of nonlinearity.

In short, we better make sure that what we remove is noise, and only noise (a random signal). This maybe helps understanding why a -60dB noise floor is already perfectly dithering the signal for many target formats.

Last edited by FabienTDR; 10th March 2017 at 04:13 PM..
Old 10th March 2017
  #385
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
It's easy to analyze via FFT analyzer. Just look for "things" building up between the partials.
Hmm, having a very hard time seeing this?

Quote:
Originally Posted by FabienTDR View Post
Be honest: Can you remember the last record you didn't like because of its quantization distortion?! I can't. if anything, I love the sound of 90's hip hop. Made with good old 16bit or 12bit samplers.
Er, lots of recordings have all manor of artifacts that are irritating to me...

Hip-hop using material sampled off vinyl into a 12-bit sampler isn't what I'd call a recipe for audio perfection. :-(
Old 10th March 2017
  #386
Sure, I'll be happy post some plots later, discussing results one by one. But I'd really prefer if ppl would try this out on their own. It's more neutral

For now, you just have to load up a FFT analyzer, set the window size as high as possible and use strong windowing. Set the floor value very low, -180 or so.

Then, simply scan through the files. Start with the original. It contains 5 segments, let's try to preserve them:

1. Stereo dirac (irrelevant to this discussion)
2. A sine pair
3. Seven sines
4. A sine pair, with natural noise floor.
5. Seven sines, with natural noise floor.

The other two files suffixed "truncatedTo16" or "truncatedTo24" have simply been exported, without any explicit dithering.

Scan through the 16bit variant and you'll find obvious quantization distortion in segment 2). But in the others, it's much harder. When looking closely, should find some quantization distortion at high frequencies, where the natural brown noise floor gets out of breath.
In segment 3) and 5) we see nothing indeed (in perfect accordance to my argumentation in this thread).

In the 24bit variant, quantization distortion disappears for all segments except 2). Interestingly, this is the most unnatural, "laboratory condition" signal.

Last edited by FabienTDR; 10th March 2017 at 04:50 PM.. Reason: fixed some typos
Old 10th March 2017
  #387
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by linkstate View Post
Try it on master and listen for space esp. in quality reverbs. You will not hear noise but slightly better separation of sounds. Almost as without dithering it smears or glues more the individual sounds.
All these things according the theory, in my point of view:

Reverberations have low level, as rule. Hence lesser signal/noise ratio. Dither increase signal/noise ratio by decreasing noise. Thus reberberations, as weak signals, is one of the first things, that dithering impact.
Old 10th March 2017
  #388
This discussion is degenerating. Permit me to contribute some technical information that seems not to be understood here. Not just any noise makes good dither. Ideally, it should have very specific statistical properties that are defined as the characteristic function having periodic zeros with correct spacing. That's what it takes to make the statistical moments of the quantization error uncorrelated with the input signal. No other noise will do that perfectly, but non-ideal noise can lower the input->error correlation and higher amplitude noise will usually lower the correlation. Sometimes that's good enough, but it depends on the input material and the noise statistics, not to mention one's experience and hearing accuity. It's totally believeable that chrism can hear a problem that FabienTDR doesn't, because each is using different source material and different noise.

So if you're willing to listen carefully on each and every occasion, and trust that no one (including a future you) will hear something you don't hear today, go ahead and substitute sub-optimal noise for proper dither. But if you'd rather spend less time on difficult listening tests and more time working on the music, then just follow standard dithering procedures and get on with it!

David L. Rick
Seventh String Recording
Old 10th March 2017
  #389
It's easy to verify whether a truncation produces partials of not. No hidden voodoo or listening in this. You calculate the autocorrelation of the removed bits, if it shows regularities, it quantized. If it doesn't (autocorrelation only contains one spike at t=0), then it's purely random and cannot, by its very nature, produce any partials.

The statistical properties of this noise only matter in the edge case trying to optimize SNR (i.e. to get it done with only 2LSB). As soon you have only 3LSB, any noise will do the job. Just try it. I did above in my tests. It's also very logical. Again, try to distort noise!

The quantization distortion manifestates as soon a periodic signal is being truncated.
With random signals (noise), nothing happens!

That's why I asked ChrisJ to truncate noise without explicit dithering and post the results. A trick question of course.

Noise doesn't distort, it cannot. That's why it works as dither in the first place! It's immune to nonlinearity. We make sure to only truncate/waste random information. The whole magic of dither. Every random noise is capable as a dither. Certain statistical properties help keeping the added noise low, while still functioning as a dither but that's it. Randomness is the key, and the fact that it can't change randomness, not matter the process it is subjected to.

IMHO, most music signals are random enough for most export situations (maybe a good reason why RME doesn't dither at all?).
No explicit dithering doesn't hurt either in average, but it is clearly more transparent, more efficient. Minimal signal path you know
I am of course aware of the provocative effect this statements has, but rethinking and verifying old habits certainly doesn't hurt anybody. And the fact that it provokes so much certainly has not a technical reason!

Last edited by FabienTDR; 10th March 2017 at 05:47 PM.. Reason: sorry for all the edits
Old 10th March 2017
  #390
Lives for gear
 
JP__'s Avatar
 

Quote:
Originally Posted by FabienTDR View Post
But I demonstrated the opposite above.
Did you use the same chain than me for this test...? No? So you cant prove me wrong either... Wir drehen uns im Kreis, und ich halte mich jetzt raus.
What David has written above is right.
You are in search for the absolute prove, for me the relative prove is more than enough in those cases. Call it fake news if this helps you to get over it..
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump