Is there a way to objectively measure dither afterwards?

[Tomás Mulcahy]

I cannot figure this out. When you only have the released mastered recording, is there any way to objectively test if dither was used, and even what kind? Seems like it should be possible but as soon as I think about it my brain turns to mush. Can you measure or track statistically the LSB and make an educated guess if it is "demonstrating" dither, so to speak?

[Hugh Robjohns]

I cannot figure this out. When you only have the released mastered recording, is there any way to objectively test if dither was used, and even what kind?

Potentially, yes... but it's difficult and potentially risky!

Realistically, the benefit of dither is only going to be audible in the quietest parts of the audio, most obviously during a slow fade-out (if such a thing exists within the material). You usually can't tell within the track itself, even during pauses, because most tracks include acoustic ambient or analogue recording noise much higher than any dither.

If dither is absent, then the last moments of the fade out will become distorted and break up... but we're talking of signals below -93dBFS which will need a lot of gain to hear... and therefore there's a risk of blasting your ears off if the next tracks jumps in while you're listening!

As for the type of dither, that's a little easier in some ways. A check of the spectrum will reveal any high HF peaks of shaped dither noise, with some named types (like Apogee's UV22, or Prism's SNS4) having very characteristic spectra.

Can you measure or track statistically the LSB and make an educated guess if it is "demonstrating" dither, so to speak?

Not really, for the reasons above. But you could try editing out the dying fade out of a track and boosting the gain digitally, or monitoring the LSBs to check the behaviour.

Be aware that most mastered CDs insert digital silence (often without dither) between tracks.

I have a CD called Cinemagic which I know has no dither anywhere on it as I was told so by the engineer who mastered it using an Neve DTC digital mastering console that had no dither! ...But I can't hear any ill effects of its absence whatsoever, despite trying!

[Tomás Mulcahy]

I have a CD called Cinemagic which I know has no dither anywhere on it as I was told so by the engineer who mastered it using an Neve DTC digital mastering console that had no dither! ...But I can't hear any ill effects of its absence whatsoever, despite trying!

This is, oddly, reassuring! Thank you.

[James Perrett]

Bob Katz designed his Bitscope to do things like this. There is also Bitter by Schwa from the Reaper team

https://www.stillwellaudio.com/plugins/bitter/

which does something similar in a VST plug-in.

[Hugh Robjohns]

RME's Digicheck tool has a bitscope too, I think, for RME interface users.

[Martin Walker]

If dither is absent, then the last moments of the fade out will become distorted and break up... but we're talking of signals below -93dBFS which will need a lot of gain to hear... and therefore there's a risk of blasting your ears off if the next tracks jumps in while you're listening!

As for the type of dither, that's a little easier in some ways. A check of the spectrum will reveal any high HF peaks of shaped dither noise, with some named types (like Apogee's UV22, or Prism's SNS4) having very characteristic spectra.

There's a developer currently beseiging various forums with demos of a new dither plugin that's claimed to be demonstably superior to the competition.

(I haven't put its name here, as in the six weeks since its initial release it's already been updated four times, new audio demos are being posted every three days on every one of these forums, while its annual rental pricing is causing huge controversy, with argumentative threads already running to nine pages in some places!)

The main reason I've posted this here is that as Hugh has already said, you can only reliably hear dither when signal levels are really low, so with typical 24-bit audio levels being dithered down to 16-bit, differences between noise-shaped dithers must surely only be audible to those with bat-like ears.

Over the years I've done my own tests to determine my own favourite dither from a selection, auditioning files with very low audio levels so the various flavours become more obvious. However, with tracks at more normal levels I simply can't hear any difference between different dither flavours, while others (presumably with younger ears than mine) do claim to hear such differences.

Can anyone else hear a difference with dithered audio files at typical levels, and also, how and why did you choose your favourite dither?

Martin

[James Perrett]

However, with tracks at more normal levels I simply can't hear any difference between different dither flavours, while others (presumably with younger ears than mine) do claim to hear such differences.

I think the bigger issue is probably when multiple stages of dither/truncation are involved. In the early days Protools offered both a dithered mixer and an undithered mixer with the idea being (I think) that the undithered mixer allowed slightly more processing power for other things. People certainly claimed to be able to hear a difference between the two.

[SafeandSound Mastering]

I cannot say I have concerned overly about which dither I have used in the past having applied MBIT+ and Apogee, and that built into any given limiter. There are just so many potential sonic elements that are magnitudes of order more significant.

I am always 100pct aware of its need and apply it and understand the principle of effectively extending dynamic range of 16 bit.

However compared to, mono incompatibility, sibilance, clicks, unintended distortion, excess reverb tails, overly wide hole in the middle (mid channel) mixes, tonal problems, occasional sounds mixed to burn your ear drums out etc. and the list goes on.

As mentioned it is for the very quietest of audio and in the vast majority of cases the analogue noise floor is higher than the dither. Though possibly not with all in the box electronic music.

In the past I have tested with and without towards ends of tracks / fades and could not hear the difference with or without at any sensible volumes/gain settings. (most likely noise was still in excess of dither)

I still deliver a lot of 16 bit masters but more 24 bit masters are requested meaning dither is even less of a requirement.

I have never favoured one over the other and given the low level of noise added (and often noise shaped for greater discretion) if the sound is changing maybe the dither source code is broken.

I am working on 90's DAT recordings the last few days and the noise floor's are much higher than any dither would be, off an analogue project console. (they are 16bit delivered to me and going back out.)

The last time I used a bit meter must have been 6 years ago or longer.

There many more important things to give attention to.

Not to mention some masters barely have 5dB dynamic range over the main course of the music never mind 90dB. Though it is not to say they are not faded out.

[Drew Stephenson]

When I still had hearing up to 14.5k (pre-covid) I did try some different dithers when I was first reading up about it. I could never reliably hear the difference between dither and no dither, let alone different types of dither. Now I've lost another k off the top I think there's even less likelihood of noticing!

[Arpangel]

If dither is absent, then the last moments of the fade out will become distorted and break up... but we're talking of signals below -93dBFS which will need a lot of gain to hear...

I remember my old Casio DA7, you could hear digital break up well before that, you had to be very careful on fade outs and reverb tails, the failings of early digital, cheap digital!

[SafeandSound Mastering]

Whether I master something digitally or with an analogue pass it is likely the noise on even a purely digitally created source with a potential theoretical dynamic range of 144dB is likely to have a much greater noise floor. The track I have just rebounced has one of approx -90dBFS according to meters (post limiting gain.)

Quite probable the source audio has noise greater than that (given it has plentiful analogue synths).

Just a few meticulously modelled analogue EQ's and compressors have this built in before it even has dither.

So many other aspects of greater concern.

[Ian Shepherd]

Can anyone else hear a difference with dithered audio files at typical levels, and also, how and why did you choose your favourite dither?

I've always thought I can (16-bit), and tried to test myself using an ABX app years ago. It was very time-consuming and incredibly frustrating and eventually I gave up, deciding that I must be fooling myself.

Then a while later testing something else I realised I'd misunderstood how the app worked, and that my testing procedure was never going to be effective with something that subtle ! So I've gone back to thinking I can, without having tested to prove it to myself.

(For anyone interested, my theory about WHY it could potentially be audible despite the fact that the distortion components are so low, is that they are correlated with the music signal (not random). The theory goes that the IMD gets so complex it becomes noise-like, which may well be true, but by my thinking it's heavily coloured shaped noise (depending on the music signal that generated it) which fluctuates constantly (again depending on the music signal), as opposed to dither which is very bland-sounding and independent from the music. Truncation distortion is kind of like adding a very low-level ugly gated noise signal.

One of these days I'll get back to it and test myself again properly, but in the meantime it's easier to just use dither !

As far as choosing a type goes, I think that's a red herring. I've never understood why anyone would master without dither and then add it later, potentially (slightly) influencing the sound. I master with 16-bit dither in place as I work, so all my sonic choices are made with that choice baked in. This removes the guesswork and then I can change to 24-bit for a (maybe) even cleaner result in the high-spec file.

Having said that, fwiw I use the MBIT+ dither in Wavelab with medium noise-shaping.

And having said all of that, most people want 24-bit files these days anyway in which case I just use simple triangular and forget about it !

(Bet you wish you'd never asked, Martin !

[Martin Walker]

(Bet you wish you'd never asked, Martin !

Not at all Ian!

I asked my question because there seem to be a lot of dither options available to us nowadays, including both more traditional flat and noise-shaped options, plus various more mathematical constructs such as Dark from Chris Johnson of Airwindows (Dark is a wordlength reducer that claims to give your music a blacker backdrop!)

Now all these 'shaped' dithers can potentially change the tonal balance, albeit at a miniscule level, so I was intrigued as to how and why people made their choices, given that the results are so incredibly subtle!

You make a very good point with your red herring comment, because you master with 16-bit dither in place as you work, such that any audible difference due to your choice of dither will be compensated for by your subsequent mastering decisions. This makes perfect sense.

However, I save my final Reaper mixes at 32-bit float and then perform various final tweaks within Wavelab before using PSP Xenon to bring the final level up to -14.0 LUFS and dither down to 16-bit. So, my choice of dither at this final stage could potentially change the sound of my mix - hence my interest in how people make their dither choices.

Martin

[RichardT]

I’ve never done a proper test to see if I can distinguish dithered from non-dithered renders. I’ve never noticed anything, and I suspect with 24 bit output files I would no chance of detecting it.

[ken long]

If you have dithered and non-dithered files of your audio, you can always sum and difference them. That should leave you with the dither to inspect. A bit meter will show you what's going on at that point but it's so negligible at those levels. I remember the Rosetta 24-Bit/16-Bit switch automatically adding the dither - it meant digital transfers would pad out the remaining 8 bits with zeros but it was a long time ago and I may be misremembering!

[Ian Shepherd]

You make a very good point with your red herring comment, because you master with 16-bit dither in place as you work, such that any audible difference due to your choice of dither will be compensated for by your subsequent mastering decisions. This makes perfect sense.

However, I save my final Reaper mixes at 32-bit float and then perform various final tweaks within Wavelab before using PSP Xenon to bring the final level up to -14.0 LUFS and dither down to 16-bit. So, my choice of dither at this final stage could potentially change the sound of my mix - hence my interest in how people make their dither choices.

Absolutely - couldn't you do the same thing, though ? At the point where you're tweaking, have a dither instance running in the chain (after Xenon) and make the final loudness choices and any other tweaks in that context ?

[NéstorRC]

Honestly, there’s no foolproof way to prove dither was applied just from the final 16‑bit master since the whole point is to hide quantization noise in a gentle haze of randomness. You can drop the file into a spectral analyzer, crank the sensitivity, and inspect the noise floor for a uniform, random‑looking smear instead of harsh quantization steps. Some engineers even plot LSB amplitude histograms to see if the least significant bits are truly random rather than stuck at zero. If you spot a subtle tilt in the noise floor, that might hint at noise‑shaping dither, but pinning down the exact algorithm is basically impossible without the original stems. In practice, if your 24‑bit mix was bounced at 16‑bit and you only hear silky‑smooth, ultra‑low‑level noise, you can bet dither did its job, even if you can’t scientifically prove every detail.

[Hugh Robjohns]

Honestly, there’s no foolproof way to prove dither was applied just from the final 16‑bit master since the whole point is to hide quantization noise in a gentle haze of randomness.

Small technical point... Dither doesn't 'hide' or 'obscure' quantisation distortion. It actually makes it go away completely.

Quantisation is a non-linear process. Adding the correct amount of dither completely linearises that process, and thus prevents the creation of distortion in the first place.

The downside is that it also introduces a fixed and constant noise floor, but that noise is (a) very low anyway, and (b) can be noise-shaped to make it perceptually even quieter.

The proof that quantisation distortion is not 'hidden' is that the peak level of added dither is lower than the peak level of the quantisation distortion products that would be there without dither.

It's a small but, I think, important point, as many people think dither is a 'fudge' to hide a problem, when it actually cures the problem and is therefore an essential part of all digital signal processing and A-D conversion.

[ajay_m]

I'm struggling to imagine a mixed set of sources that included any kind of mic'd sounds that could conceivably have a natural noise floor low enough that dither could possibly affect things.
You'd have to be working with digital sources solely in the digital domain (e.g synths) surely to be able to create a signal that didn't have a natural noise floor due to environment noise, mic signal to noise ratio etc.
Unless for some truly weird reason the 'natural' noise floor spectrum was somehow unbalanced enough that there wasn't enough random noise in a given frequency band, it's just really hard to see how effectively adding a small artificial noise floor could make any difference at all. I mean, we're working at 24 bit resolution anyway unless we're mixing down to a CD or something, surely.
Then 99% of the listeners are consuming an MP3 bitstream which has had lossy compression added and this will mask any low-level artifacts anyway, after all that's the whole point of lossy compression.

[Hugh Robjohns]

I'm struggling to imagine a mixed set of sources that included any kind of mic'd sounds that could conceivably have a natural noise floor low enough that dither could possibly affect things.

For a straight recording, it won't. And a 24-bit A-D is easily self-dithered by the inherent electronic noise from its analogue front end.

But having made that recording, if you perform a fade out the signal will eventually fall in level to cross just a few quantising levels. In a system without dither the quantising distortion then becomes highly correlated to the audio signal (rather than pseudo-random) and thus audible and objectionable.

With dither, the signal continues to fade gracefully and analogue-like into the (dither) noise floor without any distortion.

...it's just really hard to see how effectively adding a small artificial noise floor could make any difference at all.

In practice, with 24 bit systems, you can get away with out it, and most people won't ever notice... and even nerds will have a hard time proving its absence

But it does serve a legitimate purpose within all DSP manipulation, and it's really easy to implement, so why wouldn't you?

...and even if you don't, in all likelihood the DAW will apply it anyway when it converts from 32-bit floating point to 24-bit fixed point to feed the D-A converters!