Dithering is a Mathematical Process - NOT a psychoacoustic process.

[j_j]

https://ethanwiner.com/dither.html

I will need to chat with Ethan about that. He's wrong. Test conditions must be good, however.

 

[bennetng]

I will need to chat with Ethan about that. He's wrong. Test conditions must be good, however.

I posted the the link for the listening tests, not for the arguments. Artifacts are hardly audible in 16-bit truncation unless the audio materials are deliberately crafted or in very low volume.

 

[RayDunzl]

I must have missed it -- who thought dither was psychoacoustic?

I confess I would have thought so since from what I've heard (here?) it extends the signal into what was the noise floor.

(I get to be wrong, so no problem)

 

[j_j]

I posted the the link for the listening tests, not for the arguments. Artifacts are hardly audible in 16-bit truncation unless the audio materials are deliberately crafted or in very low volume.

Solo piano in a nice hall, with a quiet listening room. Rather deadly, really.

 

[KSTR]

From an engineering POV, I like to think of noise-shaped dither being a shaped random spread-spectrum high frequency PWM N-bit sub-modulation (ha, that would be a nice mfgrs. marketing buzz phrase! ;-) The higher avaible frequency headroom (higher sample rates that 48kHz), the better this works. I feel it is important to low-pass the analog signal output from the DAC chip proper to below the onset on the noise level rise. The PWM subcontent must now be factored into the reconstruction which means the reconstruction filter has to be set at a lower frequency. Further, subsequent stages are then less prone to RF-demodulate the HF noise which is already in Long Wave / AM radio range for highest today's sample rates (like 768kHz).

 

[Hayabusa]

I posted the the link for the listening tests, not for the arguments. Artifacts are hardly audible in 16-bit truncation unless the audio materials are deliberately crafted or in very low volume.

Indeed, only with very noise free signals this could make a difference. Most recordings have enough noise in them to make dither useless. Note that one way to do dithering is just adding noise before quantisation, now thats what you get: the noise is already there.

 

[bennetng]

Solo piano in a nice hall, with a quiet listening room. Rather deadly, really.

I am not those who think non-acoustic music are non-music. Using a MIDI piano (or any other instruments) would be much easier perhaps? Of course, with a long-tailed digital reverb.

 

[xr100]

Note that one way to do dithering is just adding noise before quantisation, now thats what you get: the noise is already there.

Hmm, but what kind of noise? ;-)

If you believe that properly dithering is "snake oil" for a high percentage of programme material, it is trivial to add. In that case it can be considered a free "Monster Cable." If the signal is already sufficiently noisy, then adding dither won't make much difference to the noise floor; random noise partially cancels, so even successive dithering won't build up as much as you might imagine.

If dither is missing, then eyebrows should be raised over engineering...

 

[Hayabusa]

Hmm, but what kind of noise? ;-)

If you believe that properly dithering is "snake oil" for a high percentage of programme material, it is trivial to add. In that case it can be considered a free "Monster Cable." If the signal is already sufficiently noisy, then adding dither won't make much difference to the noise floor; random noise partially cancels, so even successive dithering won't build up as much as you might imagine.

If dither is missing, then eyebrows should be raised over engineering...

Sure, this is no reason not to do dither, I only point out that its unlikely that you would hear a difference given the noise levels already buried in the recording.

 

[Hayabusa]

Hmm, but what kind of noise? ;-)

most likely thermal noise (1/f).

 

[bennetng]

I have quite a number of songs muted before fading to -inf so I can hardly tell if they are dithered or not, even after I cropped the ending of the song and amplified it.

Just to avoid others use loudness war to hijack the thread here is the whole song:

 

[solderdude]

noise gate or careless mastering/trimming ?
I suspect the latter.

 

[bennetng]

noise gate or careless mastering/trimming ?
I suspect the latter.

I don't know and don't want to guess, some songs in the same album are edited in this way, but not all of them.

 

[xr100]

I am not those who think non-acoustic music are non-music. Using a MIDI piano (or any other instruments) would be much easier perhaps? Of course, with a long-tailed digital reverb.

That's exactly what I did in the other thread.

Piano is physically modelled, Lexicon ("Room" plug-in) reverb was used, although I actually messed up the settings so it's really more early reflections than tail.

Original file, 16-bit versions (inc. one using iZotope Ozone set to "STRONG" dither and "MAXIMUM" noise shaping) are available from the link below.

I've also included JJ's MATLAB script, which creates TPDF dithered plus 3 different truncated versions of the source file. (As is the script is set to 6 bits, but can easily be modified for other bit depths.)

https://we.tl/t-JMYKKh2bGB

[Download link expires in one week.]


As for acoustic vs. non-acoustic... it's just plain wrong to think that acoustic instruments are "better." The vast majority of them aren't too good... (the best ones aren't cheap...)

 

[j_j]

I am not those who think non-acoustic music are non-music. Using a MIDI piano (or any other instruments) would be much easier perhaps? Of course, with a long-tailed digital reverb.

Depends on the verb. Many reverbs do not have that great a resolution, and barely get to 60dB. But with a good one, that would be fine.

Amusingly when I've pointed this out before people have screamed BUT THATS NOT REAL at me, so I generally cite something from a real acoustic. A real gem in the rough in the acoustic version is when there's a substantial very-low-frequency signal in the room, and you can hear the piano tail come and go with the LF signal. Oh that's so lovely. :barf:

 

[j_j]

noise gate or careless mastering/trimming ?
I suspect the latter.

I blame both. A great deal of modern production doesn't care about anything beyond the top 20dB. This is a different subject, but I've already weighed in on that hideous practice.

Also, I won't name names, but I've checked the time/frequency characteristics of quite a few reverbs, and some of them simply chop off the tail at 60 or 65 dB down from the beginning. CHOP. Gone. Poof. No, that's not good, but it happens.

 

[DonH56]

I confess I would have thought so since from what I've heard (here?) it extends the signal into what was the noise floor.

(I get to be wrong, so no problem)

It doesn't "extend" the signal; it essentially allows it to be pulled out of the noise floor. Look at @j_j's example -- I think it is very clear what is going on. It allows you to extract signals (well) below the least-significant bit level of the quantizer. Nothing psycho about it, and it works from DC through multi-GHz transmission systems, so audio is just a tiny subset of its application. My first exposure to it, and use of it to improve signal capture, was in radar systems, where psychoacoustics does not apply.

How various types of dither sound, that's psychoacoustics, I think -- and here I may well be wrong as psychoacoustics is not my field.

HTH - Don

 

[j_j]

It doesn't "extend" the signal; it essentially allows it to be pulled out of the noise floor. Look at @j_j's example -- I think it is very clear what is going on. It allows you to extract signals (well) below the least-significant bit level of the quantizer. Nothing psycho about it, and it works from DC through multi-GHz transmission systems, so audio is just a tiny subset of its application. My first exposure to it, and use of it to improve signal capture, was in radar systems, where psychoacoustics does not apply.

How various types of dither sound, that's psychoacoustics, I think -- and here I may well be wrong as psychoacoustics is not my field.

HTH - Don

Dithering, by that I mean TPD, (regardless of the spectrum of the dither you can have TPD, by the way) preserves information. By that I mean that it measurably preserves information in a signal that would have otherwise been lost.

Noise shaping can do the same thing, but can create idle tones, and other interesting things. Noise shaping should be dithered, and they aren't quite the same thing. Strictly speaking, noise shaping does not linearize completely, but it does let you get below the lsb.

 

[xr100]

I blame both. A great deal of modern production doesn't care about anything beyond the top 20dB. This is a different subject, but I've already weighed in on that hideous practice.

Hmm... it depends on the style of music? Once the transient and early reflections are done and well into the "sustain" stage, then a long decay from the instrument and long reverb tails "get in the way" on busy high BPM styles with heavy percussion. But I'd be interested in a broader (perceptual) consideration of extended decay stages; I can only think of "envelopment" with long reverb tails?

Also, I won't name names, but I've checked the time/frequency characteristics of quite a few reverbs, and some of them simply chop off the tail at 60 or 65 dB down from the beginning. CHOP. Gone. Poof. No, that's not good, but it happens.

In particular, early hardware digital reverbs, going back to the (Lexicon) 224 era, were highly limited by the hardware possibilities of the day. Given "recirculating" reverb structures, and no dither, that means digital catastrophes await in the tail. ;-)

More recent hardware may well be suspect, too; but there are some very high quality units, e.g. Bricasti (developed by former Lexicon engineers.)

Software, of course, can operate in float. I'll see about posting sample impulse responses of a few later. Some of these software processes are hardware ports and seem to suffer from "odd" behaviour.

 

[pozz]

I blame both. A great deal of modern production doesn't care about anything beyond the top 20dB. This is a different subject, but I've already weighed in on that hideous practice.

Also, I won't name names, but I've checked the time/frequency characteristics of quite a few reverbs, and some of them simply chop off the tail at 60 or 65 dB down from the beginning. CHOP. Gone. Poof. No, that's not good, but it happens.

Hmm... it depends on the style of music? Once the transient and early reflections are done and well into the "sustain" stage, then a long decay from the instrument and long reverb tails "get in the way" on busy high BPM styles with heavy percussion. But I'd be interested in a broader (perceptual) consideration of extended decay stages; I can only think of "envelopment" with long reverb tails?



In particular, early hardware digital reverbs, going back to the (Lexicon) 224 era, were highly limited by the hardware possibilities of the day. Given "recirculating" reverb structures, and no dither, that means digital catastrophes await in the tail. ;-)

More recent hardware may well be suspect, too; but there are some very high quality units, e.g. Bricasti (developed by former Lexicon engineers.)

Software, of course, can operate in float. I'll see about posting sample impulse responses of a few later. Some of these software processes are hardware ports and seem to suffer from "odd" behaviour.

We need this kind of information public for producers. I don't know of any resource anywhere with comprehensive testing of DSP, plugins or their hardware versions.