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 11th March 2017
  #451
qwe
Lives for gear
 

I've now tried the same test in u-he Hive using the preset "HS Nineteen Eighty".

Effects all OFF.

Otherwise about the same output level as Reaktor 6.

OK, that does yield a better result...

Talking of effects, I tried Reaktor with reverb, and the result was quite a bit different...
Attached Files

Synth Truncation Test - Hive - 16bit .wav (430.7 KB, 431 views)

Old 11th March 2017
  #452
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by Yuri Korzunov View Post
Truncation is one of kinds of rounding.

For closest integer rounding quantization error should be up to 2 times lesser (6 dB), I suppose. By statstics, probably, 2-3 dB. Because half values (that lesser half of 16 bit step) are rounded correctly. Need check for more exact values.
The rms error of strict truncation versus truncation after rounding is exactly the same: Q/sqrt(12), where Q is the step size. The only difference is the mean, which for convergent rounding is 0. Strict truncation introduces a small dc bias.

In other words, the quantization error range for rounding is (-Q/2, Q/2), whereas for truncation it is (-Q, 0). Both ranges have, of course, the same magnitude.
Old 11th March 2017
  #453
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
Do you have any study at hand pointing to this? All I read agree on the opposite (16bit+ of course), but I'm interested.

Theory does anyway: The threshold of audibility is higher that dithering levels. This is VERY WELL studied, also in noisy contexts, no matter what anecdotes ppl throw around when it comes to their own dithering experience, ehm, or selling snake oil.
Well it's interesting that ESS have patents on their "HyperStream" DAC technology which reduces certain artifacts in Delta-Sigma DACs.

Here's a talk from Martin Mallison, CTO:

https://youtu.be/8Mn5PrnZV-k?t=2144


Attached is a slide of an artifact found in Delta-Sigma DACs.

According to Martin Mallison, they have done double-blind tests and some people can hear this artifact. Ironically, having worked on "solutions" to the problem, he can't!

Of course, this is his assertion.
Attached Thumbnails
Good dither practices, what are yours?-ess-non-pps-noise.png  
Old 11th March 2017
  #454
^^I think they are comparing different AD/DA implementations. No idea what they are doing in detail or what the patent describes.
Old 11th March 2017
  #455
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
^^I think they are comparing different AD/DA implementations. No idea what they are doing in detail or what the patent describes.
That's correct, but it's pertinent to a discussion on the audibility of low-level artifacts.
Old 11th March 2017
  #456
Ah, Indeed.

(btw I'll later check your files)
Old 12th March 2017
  #457
qwe
Lives for gear
 

OK, I've now tested this source signal minus the inverse of a truncated version of the source material with full mixed musical content.

Commerically released recordings from the 80s sourced from CD--truncate to 8-bit--sounds and looks more or less like white noise.

My own entirely synthetically generated content--render at 24-bit TPDF dithered--truncate to 16-bit--sort of, but not quite there.
Old 12th March 2017
  #458
Nice test setup!

We should have done this from the beginning on, it's so much more friendly and accessible than me pushing thread participants to analyze test signals

I would have expected to make the same observations. No need to bother with "full" and recorded material, but with exceptions for certain super clean types of ITB material. Say, John Cage ITB



(I'm kidding, your example is good.)

Would be nice to have a tool allowing to monitor this error for any material over various target formats. I'm sure a simple reaper script would do. While quantization distortion is rather difficult to hear inside the material, the whiteness of noise is relatively easily audible and verifiable.

Last edited by FabienTDR; 12th March 2017 at 12:31 AM..
Old 12th March 2017
  #459
Quote:
Originally Posted by qwe View Post
Commerically released recordings from the 80s sourced from CD--truncate to 8-bit--sounds and looks more or less like white noise
That goes beyond my expectations. Very interesting. 8 bit, really?
Old 12th March 2017
  #460
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
Nice test setup!

We should have done this from the beginning on, it's so much more friendly and accessible than me pushing thread participants to analyze test signals

I would have expected to make the same observations. No need to bother with "full" and recorded material, but with exceptions for certain super clean types of ITB material. Say, John Cage ITB


I go to strenous lengths to minimise unwanted artifacts--and all I had to do was John Cage with auto-muting dither!



Quote:
Originally Posted by FabienTDR View Post
(I'm kidding, your example is good.)

Would be nice to have a tool allowing to monitor this error for any material over various target formats. I'm sure a simple reaper script would do. While quantization distortion is rather difficult to hear inside the material, the whiteness of noise is relatively easily audible and verifiable.
I did knock together something in Reaktor yesterday to calculate the autocorrelation function. I assumed zero mean for the standard deviation calculation, and I'm sure I've screwed up somewhere... :-(

If I take the same 16 seconds from that 80s CD to 8 bit, the autocorrelation value (of the error signal generated as per previous posts) if it's TPDF dithered to 8 bit is 0.0175. OTOH, for truncation to 8-bit, it's 750 or so.

I have no idea what these values mean scale-wise, I just fiddled with the scaling factor to get something in the "display", but at least I got a what seems to be a reasonable result there, so hopefully my "Frankenstein" Reaktor autocorrelation calculator is sort of working.

I haven't tried Reaper scripts yet--I just installed it to get 64-bit float instead of 32-bit float in Cubase. Plus Cubase seems to have some "funky stuff" going on...
Old 12th March 2017
  #461
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
That goes beyond my expectations. Very interesting. 8 bit, really?
Well, what's the S/N ratio of, say, a DX7?
Old 12th March 2017
  #462
I don't know, maybe 50dB at best? (don't know which type of DA is uses)
Old 12th March 2017
  #463
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
I don't know, but maybe 50dB at best?
Maybe slightly better than that? Roland JX series also sucks S/N wise...

So on those 80s "synth driven" recordings, by the time the whole recording and mixing process is done there's a lot of sources of "gaussian" noise and the overall S/N at best can't be much better than 50dB? So, (a certain amount of) "self-dithering" to 8-bit seems reasonable.

Maybe it would be interesting to try it with acoustic recordings from various dates.

Quote:
Originally Posted by FabienTDR View Post
(don't know which type of DA is uses)
It's erm 10-bit I think on the DX7. But regardless of the DAC (i.e. even not playing notes, so not considering quantization noise and all sorts of garbage generated due to fixed-point processing, look-up tables and the propensity of FM synthesis to generate aliasing (50kHz sample rate), etc. etc.) the noise floor is very high.
Old 12th March 2017
  #464
Acoustic recordings would be interesting to analyze indeed. If we can't hear any "music" in the error signal, and that's easy, we're safe. Given the strong dependence on mic preamps and a console, it will probably be hard to find one with a noise floor under the 2 LSB of 16bit.

I'm sure that such an example would be a much more acceptable "proof" for most thread participants than a discussion over abstract mathematical properties. I think that everybody can accept that a white truncation error is a "clean" error. It can't produce quantization distortion (that's out of doubt). It's easy to double check in any DAW.

The conclusion is slightly shocking, yes. But that should remind us to constantly question and improve our best practices.

(As mentioned several posts ago, I'm also quite surprised about these findings. I wanted to write a classy dither, so I analyzed the problem, just to find a large and depressing, but well oiled space filled by void!).

Last edited by FabienTDR; 12th March 2017 at 02:34 AM..
Old 12th March 2017
  #465
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
Acoustic recordings would be interesting to analyze indeed. If we can't hear any "music" in the error signal, and that's easy, we're safe. Given the strong dependence on mic preamps and a console, it will probably be hard to find one with a noise floor under the 2 LSB of 16bit.
It's been "known" for ages--the "skeptical" brigade have long talked about "self-dither" being the case for CD-DA, even where no dither was "explicitly" used, i.e. early CDs/digital audio before anyone was thinking much about dither.

Quote:
Originally Posted by FabienTDR View Post
I'm sure that such an example would be a much more acceptable "proof" for most thread participants than a discussion over abstract mathematical properties. I think that everybody can accept that a white truncation error is a "clean" error. It can't produce quantization distortion (that's out of doubt). It's easy to double check in any DAW.

The conclusion is slightly shocking, yes. But that should remind us to constantly question and improve our best practices.

(As mentioned several posts ago, I'm also quite surprised about these findings. I wanted to write a classy dither, so I analyzed the problem, just to find a large and depressing, but well oiled space filled by void!).

I do find a lot of this stuff absurd--on the one hand, people talk about using impulse responses taken from some old reverb box, the next about the apparent minutae between dither types.

From my POV, the "use" of this thread isn't about dither so much as gaining insight and a better understanding of various processes, which generalises to other matters.

Practically speaking, I'll continue to use flat TPDF dither for both 16-bit and 24-bit. If it's not doing anything, no harm done. It's probably consuming few less CPU than the "animated emoticons" next to the box I'm typing in.
Old 12th March 2017
  #466
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by FabienTDR View Post
The conclusion is slightly shocking, yes. But that should remind us to constantly question and improve our best practices.
(As mentioned several posts ago, I'm also quite surprised about these findings. I wanted to write a classy dither, so I analyzed the problem, just to find a large and depressing, but well oiled space filled by void!).
If your findings are telling you that people can't tell the difference between 24 bit and undithered 8 bit because given normal source material your 'quantized remainder' is indistinguishable from noise, it might be a tip-off that you're doing something wrong.

It is NOT hard to tell those things apart, and you're over-thinking it.
Old 12th March 2017
  #467
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by bogosort View Post
There's nothing "legacy" about truncation. Again, we must be clear on what we're talking about. Converters deal with fixed-point integers, so intermediate floating-point numbers must be converted to fixed-point before being sent to the converter. Likewise, delivery formats (WAV) are usually fixed-point. Dither is not required when converting from floating- to fixed-point, only when reducing the word length of a fixed-point number.
Converting float point to integer is not truncation of length because there types of data have incompatible binary structure.

Dither required for conversion any float point to 16 bit integer, because rounding errors are decreased - non-linear distortions are smoothed.
Conversion float point and 24- and more bits are similar by point of view dither applying.

These things obviously visible at spectrum of 16 bit target signal with and without dithering.
Musical signal is not suitable for testing, because there music mixed with distortions.
Need check via sine.

Quote:
Originally Posted by bogosort View Post
The only sensible way to reduce an N-bit word to an (N-M)-bit word is to truncate the last M bits. Physically (electrically), this is done by right-shifting the word by M bits.
Yes. It is simplest way, that easy applied in area of limited computing resources.

Quote:
Originally Posted by bogosort View Post
Rounding can be performed before truncation by, e.g., adding (1 << (M-1)) to the N-bit word. This reduces bias at the new LSB.
Rounding always performed before removing M bits.

Quote:
Originally Posted by bogosort View Post
However, rounding is unnecessary if we're dithering the N-bit word with, say, M+2 bits of noise: we're overwriting the rounded bit, so why bother?
Dither applied inside M bits to N bit word.

Main purpose of rounding to closest integer is minimisation of error.


For minimisation of quantization error, rounded value should be maximally close to original value.

1. Example of rounding to closest:

24bit: 0x0001FF => 16 bit 0x0002 Error: 0x000001. It is about -138 dB to 0x7FFFFF (maximal amplitude - 0 dB LFSU - of original value)

2. Example of rounding to lower (truncation lower 8 bits):

24bit: 0x0001FF => 16 bit 0x0001 Error: 0x0000FF. It is about -90 dB to 0x7FFFFF

What error is bigger?

Of course, it is emergency case and average difference will lesser, because error values distributed in range 0x00 to 0xFF. For normal distribution (I don’t know real distribution of error probability), I expect 3 times lesser - 0xFF/3 > -99 dB.
Old 12th March 2017
  #468
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by bogosort View Post
The rms error of strict truncation versus truncation after rounding is exactly the same: Q/sqrt(12), where Q is the step size. The only difference is the mean, which for convergent rounding is 0. Strict truncation introduces a small dc bias.

In other words, the quantization error range for rounding is (-Q/2, Q/2), whereas for truncation it is (-Q, 0). Both ranges have, of course, the same magnitude.
Yes. There will DC bias. It is сonsequence of simplified rounding.

DC bias give increasing error (on constant) to original value (see previous post).

If we have limitation in computing resources (on FPGA, as example), we can allow such things.

But for software implementations (resources for rounding to closest value is not matter) better way provide maximal precision without any DC, in my opinion.
Old 12th March 2017
  #469
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by FabienTDR View Post
Do you have any study at hand pointing to this? All I read agree on the opposite (16bit+ of course), but I'm interested.

Theory does anyway: The threshold of audibility is higher than dithering levels. This is VERY WELL studied, also in noisy contexts, no matter what anecdotes ppl throw around when it comes to their own dithering experience, ehm, or selling snake oil.
Unfortunatelly, I havn't information about audio tests of dither, that I have 100% sureness.

Threshold of audibility it is main unclear things in audio.

Threshold of audibility is first reason of many disputes: ringing, dither, mp3 vs. PCM, etc.

We can measure all. But interpretation of measurements stumbled by threshold of audibility.



Double blind test is type of measurements, that have number of subtlest details.

From minimal number of participants and checks to sitting places of listeners.

May be some AES members done such tests for dithering.



As developer audio algorithms, I have some experience by customer feedback about links of sound quality and measurements. I see to feedback of other manufacturers. Try link the feedback to its parameters (if available).
But it is approximate knowledges. It may be used in work, but these knowledges are not enough for official claims.
Old 12th March 2017
  #470
qwe
Lives for gear
 

Quote:
Originally Posted by chrisj View Post
Quote:
Originally Posted by FabienTDR View Post
The conclusion is slightly shocking, yes. But that should remind us to constantly question and improve our best practices.
(As mentioned several posts ago, I'm also quite surprised about these findings. I wanted to write a classy dither, so I analyzed the problem, just to find a large and depressing, but well oiled space filled by void!).
If your findings are telling you that people can't tell the difference between 24 bit and undithered 8 bit because given normal source material your 'quantized remainder' is indistinguishable from noise, it might be a tip-off that you're doing something wrong.

It is NOT hard to tell those things apart, and you're over-thinking it.
I did three tests: (1) With a "clean" synth (Reaktor 6 4x oversampled) truncated to 16-bit, (2) "Wavetable" synth (u-he Hive) truncated to 16-bit and (3) Synth-driven music from a 1980s commerically released CD.

(1) and (2) were truncated to 16-bits. The source signals for (1) and (2) were at about -53dBr peak, so obviously, not all 16-bits were used when truncated.

(3) was truncated to 8-bits.

Then, the difference between the original and truncated signals was output.

The test "failed" in the case of (1), in (2) was noise-like (the sound had a slow release time, so the quantization error was quite audible at the end), (3) noise-like (again until the fade out.)

So the results of these tests do NOT imply that 8-bit truncation sounds the same as the source under any of these conditions, only that it's not always the trainwreck that one might anticipate.

You can listen to the outputs of (1) and (2) (normalised to -3dBr) for yourself...

https://www.gearslutz.com/board/atta...test-16bit.wav
https://www.gearslutz.com/board/atta...ive-16bit-.wav

...and deduce whether or not the processing is incorrect?

I would be delighted if I had made a mistake which could then be rectified.

BTW, your "NotJustAnotherCD" plugin (with the same "quantization error" test) is producing most odd results when fed with a sine wave at low levels?

EDIT: Based on a quick test in Reaktor, it produces the same artifacts as quantization with the quantization error subtracted from the next sample. In fact this was used in the early Philips chipset-based CD players, with 4x oversampling, in an attempt to get 16-bit performance out of the Philips TDA1540 (14-bit DAC.)
Old 12th March 2017
  #471
How did the 16bit and 24bit test perform?!

8bit doesn't really follow modern standards, a premise I clearly communicated upfront. If it works, it will be because of an insane noise floor (say, a recording from the 60s).

The main question is, can we detect any problems with music material in 16bit and 24bit? (the removed content not being purely white). IMHO the answer is: Typically NOT, but with rare exceptions for unnaturally low noise floors. Redbook content is 100% safe from any trouble to 16bit, as it naturally contains at least the redbook noisefloor.

It is very difficult to provoke any quantization distortion in modern music production workflows (32bit fp monitored as 24bit fixed). A noise floor higher than your needed LSB is almost always given, if you're not just playing with 2-4 sine.

Any material that has been recorded will be immune over a large range (definitely down to 12bit).
Explicit dithering shouldn't become a paranoid dogma! It's lazy, wasteful and pointless in the majority of the cases.

Nobody could ever demonstrate ABX audibility of 16bit truncation and 24bit truncation of music material, it's 2017! This is quite interesting seeing the enormous hype behind dithering and its workflow police. A dogma like many others.

No dithering study on the whole internet considers the effects of naturally noisy signals or the effect on complex material (more than 2-4 sine), they are literally ignored and only sines are discussed! That's weak. Super weak. Music always contains noise, so it would be interesting to know what complex, noisy signals do in this context. But nobody ever did (feel free to correct me). I have a suspicion why It radically questions the "written in stone and never really verified" practice of dithering music in hi res context.

The lack of practical motivation in certain academic and commercial circles is apparent, and most aren't into music signals anyway. Outside lofi cases, its effectiveness is very hard to demonstrate. Lol, obviously even for ppl building and offering such products!

Given this:
https://en.wikipedia.org/wiki/List_of_cognitive_biases
in the context of this (Yuri, this IS well studied):
https://en.wikipedia.org/wiki/Absolu...old_of_hearing
Personal anecdotes can be clearly thrown into trash. Nothing easier than making things up in music, especially at the threshold of hearing (which is officially the point were ppl "make things up"). The ear worm is a very real phenomenon, there's even clinics dedicated to them. When even my grandma can imagine a whole song in her mind, a pro will certainly do even "better", especially if he's been told that this thing that's officially too low in level to be perceived, and he doesn't understand its mechanisms anyway, greatly improves the audio.

Next time someone goes crazy because you didn't dither, safely ignore him and be aware that your decision is technically well funded. Much better funded than dithering of noisy signals itself. The noise floor of most signals is ridiculously higher than 2LSB (dog house vs sky scrapers here). The benefits of dithering in modern music production are pure belief. Practical benefits can't be shown, except under (very restricted) laboratory conditions. This also explains well the "magical" audiophile vocabulary being used to describe the sonic advantages of this process.


Last edited by FabienTDR; 16th March 2017 at 07:33 AM..
Old 12th March 2017
  #472
IMHO, the discussion about explicit dithering in modern audio production contexts is a good example of Parkinson's law of triviality.

We talk about LSB and forget that noisefloors are way higher anyway, our perception is physically limited and music is never sine. Three big fat elephants (compared to one fly in terms of scale!) that get rarely mentioned or involved in these discussions. Classic bike-shedding.
Old 12th March 2017
  #473
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
IMHO, the discussion about explicit dithering in modern audio production contexts is a good example of Parkinson's law of triviality.
Hmm, that actually explains a lot--why there are discussions on dither and and also matters such as using an "emulation" effect to replicate some awful old microphone that belongs in the dustbin... but ignoring the really (and sometimes much harder to grasp) important stuff.

(Obviously emulating a microphone properly isn't so straightforward, but the basic premise--"this will make it sound like the material was captured using [famous microphone XYZ] used by [famous engineer ABC]"--is.)
Old 12th March 2017
  #474
Indeed.

The main issue for me personally is not the fact that ppl are using explicit dithering in their workflows.

It's when they recommend/enforce it publicly, without really being able to prove much relevance beyond lofi cases (12bit and below) demonstrated under specific laboratory conditions (noise floors below -100dB, very simple signals).

Explicit dithering has its uses, that's out of question, but not too many in this modern context. It's well worth communicating that the benefit really isn't clear at all. If you like it, it's fine, but it really handles a very academic issue, without any benefit in the majority of the cases.
Old 12th March 2017
  #475
Lives for gear
 
stinkyfingers's Avatar
 

i find it funny that the concern for dither is so high but the concern for low resolution automation is non existent. but it is out there and people use it every day and don't notice, or just don't care. IMO automation is a key tool and it needs to be of good precision/quality.
"stepped" automation is bad news. and i don't think dither can "fix" it.
it was pretty easy to get a good amount of artifacts to pop right up out of the noise floor on the 7 sines test w/noise by using coarse automation...
it seems to work the same with music as well, not just sine waves.
and it seems wideband material doesn't just melt into white noise like truncation, either. i haven't tested with any full blown orchestral pieces or black metal or anything good, just mostly acoustic and synth stuff.
trying to add dither pre/post the automation only makes it worse.
so any DAW or plug in that has "low resolution" automation is going to be doing a lot more damage than rounding/truncation to 24 bit.
Logic had/has like 8 bit fades (or is it 7?)...8 bit ? and people aren't noticing this ? you're kidding me. it's got to be because everyone just slams the **** out of everything anyways so there's no room for any more noise. i know that "masking" has great powers, but surely not all the time every time.
i guess i have to agree with Fabien now about "noise will be noise" (bad noise, bad noise). not that i ever disagreed, i just took offense to the derogatory use of the term "noise", as a "noise artist".
i heard Logic's **** automation problem the first time i ever used it (tried to put a slow fade on a synth track.) never used it again. it's terrible.
you want to turn a silky smooth cello track into a fluttering dog fart ? automate it in Logic.
sorry for the rant. maybe I'm just used to the nice fuzzy scratchy crackly dog cat hair smoke ash coffee gummy bears sticky faders on an analog mixer.

Last edited by stinkyfingers; 12th March 2017 at 08:51 PM..
Old 12th March 2017
  #476
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by FabienTDR View Post
How did the 16bit and 24bit test perform?!
Comparing 16 and 24 bit technically impossibly.

Because DAC work in different modes and result of comparison depend on implementation of these modes in general. Ant there are compared modes of DAC, not resolution. For such test need exactly (very exactly) know how to work tested DAC.

Comparing 16 bit with dither and without is possible.

For testing need use low level signals (-40 dB and below for first approach). Also need check different SPL levels.
Old 12th March 2017
  #477
qwe
Lives for gear
 

Quote:
Originally Posted by Yuri Korzunov View Post
Comparing 16 and 24 bit technically impossibly.

Because DAC work in different modes and result of comparison depend on implementation of these modes in general. Ant there are compared modes of DAC, not resolution. For such test need exactly (very exactly) know how to work tested DAC.
I'm confused--if you're sending the DAC 16-bit data as a 24-bit stream, then it's not switching mode?

Or do you mean that, by zeroing out the 8 LSBs it will have an impact on the DAC's operation like delta-sigma noise shaping loops?
Old 12th March 2017
  #478
qwe
Lives for gear
 

Quote:
Originally Posted by FabienTDR View Post
Indeed.

The main issue for me personally is not the fact that ppl are using explicit dithering in their workflows.

It's when they recommend/enforce it publicly, without really being able to prove much relevance beyond lofi cases (12bit and below) demonstrated under specific laboratory conditions (noise floors below -100dB, very simple signals).
The trouble is what's happening once the signal is either processed again or what happens in the DAC?

That's one of the reasons I posted: https://www.gearslutz.com/board/12497862-post452.html

Some of these problems are very difficult to find and measure, according to the talk linked from that message, the only way they can fully characterise it is by running simulations looking at multi-dimensional state space. Added to which not everyone can hear the artifacts in a double-blind ABX test. (They have FPGA implementations of their DACs where they can turn on/off different features.)

Quote:
Originally Posted by FabienTDR View Post
Explicit dithering has its uses, that's out of question, but not too many in this modern context. It's well worth communicating that the benefit really isn't clear at all. If you like it, it's fine, but it really handles a very academic issue, without any benefit in the majority of the cases.
If your goal is the very best possible sound quality, then you've got to look at all details, big and small. This means, for instance, you can't go using some canned drum loop--the provenance of which is likely unknown and may well breach the extreme criteria required.

This may be a quasi-academic exercise in so far as it doesn't relate to common practice, but it's not academic in that the intended output is "real music." :-)

I agree that if what you're doing is "average-fi" stuff, then it's probably not something to worry about. Just render from floating-point to 24 or 16-bit with flat TPDF dither, and don't worry about dither ever again...
Old 13th March 2017
  #479
Airwindows
 
chrisj's Avatar
Or, indeed, if you're that offended by the idea that people do critical listening and make decisions about fugitive sensations mostly associated with subtle soundstage depth cues and textural issues, render to whatever the hell you like and get out of our grills when we gather to discuss how WE explore the farthest reaches of this audio nerdage.

I am MORE than happy to spend days and weeks and months and years working out what would best condition digital audio for my Neil Young 'Pono', which is an AMAZING little portable audio player with known and ingenious D/A conversion (it does some unorthodox things given high sample rate). I'll do this knowing perfectly well that almost nobody has a Pono. Most people have iPhones or cheap Android phones and earbuds. I don't have to care about that, and I don't. The audio I (and my compatriots) produce, is there for if those people want to get all audiophile for a hobby.

Trying to get in our way is (a) ludicrous and (b) impossible. I likewise don't care how many posts you make calling me a bozo. I have a Pono (and a Motu 16A, and a Lavry Black DA-10) and you don't. And some folks who post here have way, way better than that. You can do what you like, but you're in no position to tell us NOT to care. About our ruthless audio nerdage
Old 13th March 2017
  #480
Hey, I do that too. I need my couch, some weed, great music and enjoy this kind of stuff you are describing. A lot in fact!

This an agreeable ride on your own feelings, arbitrarily enhanced by your previous musical experiences, your mood, your intentions, and and and. A good feeling.

Point is, I try to remain aware that this has little to do with the scientific method or proper engineering. While this aspect is absolutely day night obvious in any other field of engineering (medical, defense, navigation, gfx processing and video, communication in general, finance, ERP, big data, and so on), in the music production scene in particular, it is not.

Only music has the power to trigger auditive imaginary, the music experience in itself. I suppose that this is apparent to most people. This is both the reason why so many love this job so much, but also explains well why the scene has such strong tendency to generalize their own, often very vivid, but personal experiences. Music is subjective experience fireworks and that's great.

I don't question the vividness of personal music experiences, just the relevance it has for other people. I know very well how strong the effect of positive expectation can be, often to find out, afraid, that bypass was engaged over the last 15 min. So I try to stay humble regarding this kind of stuff, simply because making things up is the whole point behind music consumption. It's too easy to run into this trap, staying out of it asks for a lot of discipline and depressively boring work (yes, really totally boring. But worth the trouble).

Last edited by FabienTDR; 13th March 2017 at 02:55 AM..
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