I am experimenting with how dithering affects the sound quality when placed at different points in the mastering chain. I would appreciate your opinions.
https://pixeldrain.com/u/3Vv8jS2A

 

Usually, dither is meant to be the last in line but here they all nil.

 

The last after limiter, SRC, editing etc.
Might not work for lossy, mp3, depends on codec algo etc.

 

Only when you go from high bits to low bits. From float to integer. After all processing.
To avoid quantization noise, distortions, truncation, rounding. (Truncation and rounding are different terms, both are shit when not dithered).
TPDF is classic, but noise shaping might be better for ears, but technically may cause problems. You also can apply noise shaping without dithering, it such a way of making quantization errors psychoacoustically "masked".
Dithering is not masking

Roughly said.. Sorry, a bit drunk

 

What No Avenger and mild pump milk said. Also, the difference between different dithers is largely imperceptible. The biggest difference is using it at the end of your mastering chain, or not using it. For example, just using most common TPDF dither will make a bigger difference than not using any dither at all.

 

I remember when I rendered bass sample in Audacity WAV 16-bit with Triangular Dither the result had some harsh on the top.
Wtf? Sound had subtle artifacts.

Since this I never used Dither and I have it always Off.
And no more harsh or artifacts!

Even in your example. The last half of the second.
There is subtle more noise in "After" than "Before".

 

I do think they sound slightly different - if you deliberately test for it. Loop the very end of the fadeout and put a limiter on it with +60 db (or more).

imo the fadeout in 'Both' always sounds the worst. Not sure between 'After' vs 'Before', maybe 'After' sounds a bit cleaner and 'Before' a bit noisier - maybe.

Anyway, you have no choice but to put the dither at the very end if you want to de-correlate quantization distortion. If the dither is anywhere else, it's just extra noise and you get correlated quantization distortion at the end.

 

I understand the established theory, but I see this as a test—a 'what if' scenario. Exploring this against-the-theory technique might offer a unique creative tool. In my assessment, its most significant impact lies in altering reverberation and transient response. However, it's crucial to note that implementing any effect inevitably entails sacrificing some clarity (which is always the case when implementing an effect)...
I would appreciate hearing some more opinions.

 

My humble input: I have no idea. Blind tested, couldn't tell them apart consistently. But i'm listening in a not ideal enviroment and monitoring system so idk, maybe there's something else in there.

Here's a question for you. Null testing this thing shows nothing but VERY low level noise. The difference doesn't seem to be doing anything in terms of dynamics (you would expect to see a bit of movement if there was something affecting transients) Although tbh i don't have an analyzer fast enough to register fast stuff, if that was a factor. So the question is:

Have you measured that somehow, or is that just based on consistent listening tests?

 

For creative and transient shaping/faux stereo purposes, I'd look into velvet noise (examples towards the bottom of this link).

For dithering - as in: defeating patterns from quantization distortion to 'objectively' improve audio, I think virtually no one is going to notice the difference between any two triangular dithers.

 

LOL
Let's talk about what dithers actually are for.
So, digital audio has the lowest possible loudness, determined by bit depth. For 16 bits it's about -60 dB, for 24 bits it's about -120 dB. I'll talk about 16 bits from now on for simplicity, in case of 24 bits, 60 changes to 120. But what happens if the audio gets lower than that?
Well, if you ever turned the "depth reduction" knob on your bitcrusher, you've heard exactly what happens, but 60 dB louder. Sound gets all harsh and jittery before disappearing.
This is what happens to the most quietest parts of your audio when you bounce, for example, a 32-bit project to a 16-bit file. Now, what does dither do?
It adds an artificial, thoroughly calculated noisefloor that's exactly -60 dB loud. so no signal ever gets below -60 dB in volume, and no jittery bit reduction happens.
Obviously for this to work, the noisefloor must be unprocessed, i.e. at the very end of a signal chain, after even all the mastering.
It's too quiet for any creative use, for that you would need a proper noise generator with envelope follower or something.

 

Actually, the lowest bit is -90 dBFS at 16 bits and -138 dBFS at 24 bits.

So listening to a -14 LUFS leveled 16 bit Signal at 85 dB SPL will result in a 9 dB SPL noise floor using 1 bit for simple dithering. Now tell me you can hear the difference between different 9 db SPL noises. With noise shaping, the noise in the mid range where the hearing is most sensitive goes easily below 0 dB SPL.

On the other hand, if you are listening to a -23 LUFS orchestral recording at 97 dB SPL so that the loudest parts can blow you away with 120 dB SPL, a resolution of 16 bits would result in a 30 dB SPL noise floor without noise shaping. In this case, different dithering settings with good noise shaping should be quite audible. But only until the recording hits the 120 db SPL for one second. After that you won't hear the noise floor for hours.

 

You're probably correct about precise numbers, but I hope I carried the idea across.

 

As far as I can hear, and some members confirmed independently, there is an audible difference between all of the files posted. Each has a slightly different character.

From what I can discern, the primary impact lies in the transient and reverberation (ambient) domains, as far as perception through hearing is concerned. This also manifests in variations of darkness or brightness, as well as sharper or duller tones. If individuals can perceive these differences without any prior suggestion of what to listen for, I believe it's a genuine phenomenon. Therefore, it could be a feasible and beneficial technique to explore in order to achieve a specific sound expression. As far as I am aware, no one has previously proposed such an approach on this site.

I don't see the utility in measurements or theoretical considerations because my aim was solely to determine if people can hear the sound differences as I do – even if any differences exist at all.

 

That's almost exactly how I heard it.

 

I made a second phase inversion test and in fact they do not nil, so these are the results

Spoiler: levels...

Spoiler: ... and freq difference between two files

Have a guess what it sounds like.

 

Yeah, but consider those are also the things (bright, dull, sharp, etc.) people usually point out when comparing audio things that have too little a difference or none at all. We've seen it before when people test bit dephts, sample rates, plugin emulations, even fake comparissons when the files compared are the same. It's about the only way you can describe sounds subjectively, but not very reliable.

Well, not anyomre. and there's only two answers that claim to have heard a difference, and the two are opposites.

Not saying that there isn't a difference, because there is, following what Xorome suggested, before is indeed noisier than after and it's even measurable, so there's at least that.

 

When discussing audio perception, we depend on adjectives to convey our experience. Language remains our sole tool for articulating these insights. However, relying on adjectives doesn't imply a perception gap of "barely anything" or "nothing at all."

It is more likely that individuals may be unable to hear something due to their lack of methodology (take xorome as an example of the opposite), the low quality of their gear, lack of attention or concentration, or even due to limitations in their hearing abilities, rather than the possibility that individuals who can hear something are imagining things.

 

For me the last second in "After" have more noise.
Why?
Because this "whyp" at the end is like through compressor. It increasing noise at the very end.
The end in "After" have more noise in my opinion.
Here is this on graph presented visually for better understanding.


----------------------------------------------------------------------------------------------------------

 

You are correct because dithering involves adding noise, and its impact depends on the subsequent processing that is applied.