Dither - My Ears can't hear it

[pezking]

Ever listened to flute that doesn't have dither? It's horrific.

The whole point of dither is that it's nicer (and less annoying) to listen to noise (i.e. dither) rather than distortion (quantizing distortion).

Ethan your argument is flawed - you can't guarantee that someone doesn't want to listen to the end of that reverb tail - in classical music, you will frequently hear sounds that go right down into the +/- 1LSB area. I recorded on location before Christmas, a performance of O Magnum Mysterium. It damn near peaks in places, and in others is right down there. It'd sound horrific without dither.

Perhaps in a pop/rock track with no range of dynamics the effects of dither are masked, but listen to something with a wide range of dynamics and you'll understand why dither is necessary.

I've just made a small demo of dither using very low level sine waves. Anyone want me to share it?

 

[Ethan Winer]

Ever listened to flute that doesn't have dither? It's horrific.

Surely you exaggerate. Tell you what - email me a 24-bit file of a flute playing at normal levels peaking within, say, 6 dB of full scale. I'll post truncated and dithered versions for folks here to audition and guess which is which. There's a link to email me on my home page www.ethanwiner.com.

Ethan your argument is flawed - you can't guarantee that someone doesn't want to listen to the end of that reverb tail - in classical music, you will frequently hear sounds that go right down into the +/- 1LSB area.

Okay, let's analyze that scenario. If a piece of classical music is very loud it might hit 100 dB SPL on peaks. Dither is 90 dB below that, so that's equal to a noise floor of 10 dB SPL. What concert hall or living room has a noise floor anywhere near that low?

I've just made a small demo of dither using very low level sine waves. Anyone want me to share it?

I'd much prefer a recording of a flute or some classical piece so we can assess it in context at normal listening levels.

--Ethan

 

[masteringhouse]

Ethan -

The test file is 16 bit, so I'm not sure what it proves since from your description part was truncated, part was dithered, then the edited pieces were combined and dithered again. You basically masked out the difference with a second dither. It's an invalid test.

Here is a simple test, listen to the same file and change the type of dither. Also try taking the same file and truncate, then render again with dither. Listen back on a CD or bring it back in as 24 bit. Do you hear a difference? Many do. If not, get a better monitoring system and try again. If you feel the need to argue further speak with Dan Lavry, Nika Aldrich, Daniel Weiss, and other top DSP engineers. I truly doubt that they dither for no good reason.

 

[SouthSIDE Glen]

As soon as you test them blind all of a sudden that have no idea if the music has been dithered or not.

That's a unfair testing parameter right up there with "listen to this and tell me what samplaing rate it was sampled at."

It's not that dithering has an identifiable "sound" or color that can be isolated and identified, any more than 44.1K has an identifiable "sound". It's not possible to listen to a recorng out of context - or even within some conetxts - and say definitively whether it has been dithered or not. But that's not the point in either case.

The question is to whether dither *changes* the sound. While on paper it's easy to demonstrate that dither should never audibly change the sound, I can tell you that I have "sensed" a difference (for lack of a better term). Not every time, mind you; I find it to be very content-dependant. Could I always hear a difference? No. When I could hear a difference, could I guarantee which one was dithered and which one wasn't? Not always, but more than 50% of the time the one that had a more comfortable, almost a more "analog feel" to it was the one that had good dithering applied to it. I couldn't necessarily pin it down to a describable effect, just that it left me feeling better.

I personally don't find it a black and white issue (I know, I'm dithering myself! Oh, the irony! ) I say try it and A/B. Pick whichever one makes you feel better. If it's the dithered one fine. If not, equally fine. If you can't really hear the difference, that's just fine too. But in my limited experience with the issue there are enough times where the effect, while very subtle, is real enough in enough to take seriously and not just dismiss as voodoo.

Just one man's anecdotal experience. YMMV.

G.

 

[pipelineaudio]

SouthSIDE Glen, can you have a look at this plugin?

http://www.stillwellaudio.com/?page_id=27

heres a direct download to the evaluation verion (uncrippled) http://downloads.stillwellaudio.com/schwa/PsychoDither-install.exe if you have a VST compatible host

 

[SouthSIDE Glen]

Que es que se? Those guys keep calling "God's own dither" by the name "Phycho Dither", they better have a damn good ligtning rod!

I've d/l'd the demo version, but right now I am near a system that has only VST-via-DX-wrapper capability, and I frankly don't much care for the wrapper performance. When I get back to my bigger system I'll take a look.

But honestly, I think you're asking the wrong guy. I'd much rather hear what Tom or John have to say about it; they have far more experience with different dithering algorithms (Ozone and Audiotion are all I have worked with to any extent) on far better monitoring platfooms than I have.

BTW, I need to contact you and talk with you in a day or two regarding something totally different. Right now, though, I have a severe thunderstorm bearing down on me and I'm going to button things up untiil it passes. BBL....

G.

 

[giraffe]

do i do it? yes.
can i hear it, no, not really.

no exactly taxing on the cpu, so whatever.

 

[jrhager84]

I feel stupid. Am I wastefully recording at 24bit?

*goes and sits in corner*

 

[Ethan Winer]

You basically masked out the difference with a second dither.

I did no such thing! The truncated file is truncated, period. If this is not clear on my web page, please show me what words I should change.

Do you hear a difference? Many do.

No, they only think they do. As soon as you test them blind they can't tell anymore.

This is just like people who claim to be able to hear different speaker wires. They can identify the wires only when they can see which is which.

I applaud Matt (OP) for having the courage to stand up and say he can't hear any difference.

--Ethan

 

[Ethan Winer]

That's a unfair testing parameter right up there with "listen to this and tell me what samplaing rate it was sampled at."

I don't see why it's unfair. I switched the dither on and off right in the middle of passages! If that won't reveal a change, what better test do you propose?

on paper it's easy to demonstrate that dither should never audibly change the sound

No shit.

I couldn't necessarily pin it down to a describable effect, just that it left me feeling better.

This is called "expectation bias" because you perceive what you think you should perceive. But have someone else play dithered versus truncated and I promise you won't be able to "sense" it any more than 50-50 random chance.

--Ethan

 

[Ethan Winer]

I feel stupid. Am I wastefully recording at 24bit?

Yes, but that doesn't make you stupid! It just makes you one several million victims of a massive PR campaign to waste people's time and money on silly nonsense that has no effect on audio quality.

--Ethan

 

[pezking]

Yes, but that doesn't make you stupid! It just makes you one several million victims of a massive PR campaign to waste people's time and money on silly nonsense that has no effect on audio quality.

--Ethan

Let's not stray from the subject of dither - bit depth has been discussed elsewhere on these forums and it is universally accepted that 24bits is far better than 16 - I've conducted research that backs this up!

I've just visited your website. The opening paragraph seems to illustrate a lack of understanding of dither, and of A-D conversion.

Dither isn't just necessary when reducing the bit depth of a digital file. It's necessary every time you convert from A-D and every time you sample there after. Whether you like it or not, your converters are adding dither.

I've not had a chance to listen to your demo yet, but I'll endeavour to do so in the next hour.

 

[mattkw80]

I applaud Matt (OP) for having the courage to stand up and say he can't hear any difference.

--Ethan

Thanks Ethan !

But I didn't mean to start an argument!

I've listened to your "tv test" audio example 3 times, with my monitors cranked. I cannot hear anything changing, or happeing. To me it seems like the same file throughout.

So, I don't have Pezking's listening experience, or Bob Katz's listening experience - those guys hear far better than I can.
(Those guys probably hear about "30 years" better, beyond what I can hear...)

However - that makes me "average joe listener".

And if average joe listener can't hear it - it's in audible.

IMO...

If average joe listener can't hear it, listening to the radio, driving down the road - it's inaudible.

If average joe listener can't hear it at a club with 100 people talking over one another - it's inaudible.

If average joe listener can't hear it, sitting in his living room ,listening hi-fi on $10,000 worth of B&W speakers (which my uncle has, and I use whever I want) - then the dithering would be in-audible.

If pezking can hear it -- he's a pro, that could be expected.

Take all that - and couple if with the fact that songs are now jammed into $50 Me-pod boxes, after losing 90% of their file size and audio information......... IMHO, dither does nothing for me.


Some will argue "Well just because you can't fully appreciate the Mona Lisa's detailed brush strokes, that doesn't mean the artist should have left them out, or that more experienced people who do recognize fine detail shouldn't enjoy them".

Which would be true enough.... but I cannot hear a difference, and would guess that 95% of those here cannot.

 

[Ethan Winer]

Let's not stray from the subject of dither

I see bit depth as directly related to dither, and not that far away from ultra-high sample rates.

it is universally accepted that 24bits is far better than 16 - I've conducted research that backs this up!

I don't accept it! I've done research too.

Dither isn't just necessary when reducing the bit depth of a digital file.

That's the only type of dither I'm addressing - the kind you can use or not use when rendering a 24 bit mixdown to 16 bits to put on a CD.

I've not had a chance to listen to your demo yet, but I'll endeavour to do so in the next hour.

Excellent! Let me know what you find. Or don't find.

--Ethan

 

[Ethan Winer]

But I didn't mean to start an argument!

I never argue, I only discuss. Others like to argue though.

Some will argue "Well just because you can't fully appreciate the Mona Lisa's detailed brush strokes, that doesn't mean the artist should have left them out, or that more experienced people who do recognize fine detail shouldn't enjoy them".

Absolutely.

I cannot hear a difference, and would guess that 95% of those here cannot.

Ah, but nobody can hear dither. It's not just you. Some have convinced themselves they can hear it, but they really can't.

--Ethan

 

[pezking]

If pezking can hear it -- he's a pro, that could be expected.

I am not a pro and don't claim to be. I will be one day though

That's the only type of dither I'm addressing - the kind you can use or not use when rendering a 24 bit mixdown to 16 bits to put on a CD.

The dither is applied then for exactly the same reasons as it is at A-D conversion. It is applied every time the signal is resampled!

Ah, but nobody can hear dither. It's not just you. Some have convinced themselves they can hear it, but they really can't.

In your opinion. Your claims are dependent on the signal never being quiet enough for the listener to be able to hear the dither, so giving a blanket claim that "nobody can hear dither" is completely and utterly wrong and so far from the truth.

Next week, when I'm back in the studio I'll try to record something and demonstrate dither.

Matt, please take into context what you read on these forums - read a book, understand exactly what it is that dither does, doesn't do, and why. I think whilst these forums are a valuable tool to the home recordist - you can't take what your read hear as gospel (that goes for what anyone says - including me), and it's no substitution for traditional educational resources (books!!!).

 

[masteringhouse]

Ethan -

Your test file is 16 bit, so you either truncated the final file or dithered its content. If you dithered the file then you are dithering the truncated audio, and double dithering the dithered content. If you truncated on final output, then you are adding truncation artifacts to the dithered content and leaving the truncated content intact.

The test file is confusing the issue and isn't a valid test of anything useful that I can think of if I understand correctly what is contained in the file.

 

[SouthSIDE Glen]

I don't see why it's unfair. I switched the dither on and off right in the middle of passages! If that won't reveal a change, what better test do you propose?

It's one thing to see if a change is detected. It's another to ask someone to ID which is which. Adding dither is not like adding salt or pepper, it does not have a specific, identifiable, unique flavor. It can cause change, but the change may not be imeediately ideantifiable as "dither".

It's like switching between two preamps. One may or may not hear a difference, but to be able to say which preamp is which often not possible (except with *really* colored preamps or pres that have been pushed to coloration, but that's a different story) because it's effect upon the sound varies by mic, source sound, etc.

The test also is on one sound file using one type of dither. Because the resulting effect dither can have - positive or negative, apparent or non-apparent - is dependant upon both the dither type and the content. It's very easy to set up a single test that seems to demonstrate one given point. What's needed is a body of expirimental data, not just one spectific test.

This is called "expectation bias" because you perceive what you think you should perceive.

Not true in this case, for three reasons. One, because any particular "expectation" has been wiped clean by repeated blind experience. Two, because my initial expectation was heavily on the skeptical side. Three, because the results vary by algorithm and content. There's no "bias" because I claim no specific "bais" in the composite results. Sometikes it helps, sometimes it doesn't. Sometimes it doesn't matter at all.

I don't EQ everything that comes my way, and sometimes EQ can make things worse or not help at all. But very often, EQ is helpful. I see no difference with dither in that regard.

But have someone else play dithered versus truncated and I promise you won't be able to "sense" it any more than 50-50 random chance.

I can play you two mix versions and there's no way you could sense which one had the EQ and which one did not. There's no way you could tell me which one used mic A vs. mic B for the backup vocals. There's no way you could tell me (if I did a good job) if that snare was punched in on this take or that take. There's no way you could tell me which one was mixed on Tannoy monitors and which one was mixed on Genelecs. There's no way you could do any of those on a better than 50-50 basis.

The point is not whether one can ID what you did and did not do to a mix to make it sound that way, that's pretty meaningless. In fact, it's often just the opposite. If something *sounds* as though it's been EQ'd, compressed, or reverbed, that is often (depending upon content) a sign that the engineer messed up. Why should something sound "dithered"?

All that said, though, it's a tempest in a teacup nonetheless, compared to a dozen other issues. It's right there with sample rate; if you're worrying about either one, you're probably worrying about the wrong thing. It's like trying to get a woman into bed and worrying about whether your condoms are ribbed or not. You have a lot of more important bases to cover first.

G.

 

[Ethan Winer]

It is applied every time the signal is resampled!

Not sure what your point is, but the test file I created was made by rendering a mix to a 24 bit Wave file at 44.1 KHz. That was then loaded into Sound Forge and truncated to 16 bits and saved, then I did the same thing again using dither and saving to a new file. Only one of the files was dithered to get from 24 bits to 16 bits. The other was truncated. There was no resampling. I don't know how I can explain this more clearly!

Next week, when I'm back in the studio I'll try to record something and demonstrate dither.

Excellent. But please don't use trickery, like recording stuff at -40 then raising it before uploading etc. And don't expect us to increase the volume 30 or 40 dB to hear it. To be a valid comparison the file must contain music recorded at "normal and usual" levels.

--Ethan

 

[Robert D]

Some will argue "Well just because you can't fully appreciate the Mona Lisa's detailed brush strokes, that doesn't mean the artist should have left them out, or that more experienced people who do recognize fine detail shouldn't enjoy them".

.........or that less experienced people won't subconsciously appreciate them, even if conciously oblivious to them. I think there may be some validity to Glens premise that dither isn't, at normal listening levels, something you hear and say "listen to that fine dither", but rather something that at some less definitive level does enter our perception. It wouldn't be the only thing in the human experience that is hard to quantify.