Dither - My Ears can't hear it

[Ethan Winer]

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.

Man, I must be losing my communication skills!

The file contains portions that were truncated and portions that were dithered. This is the whole point. The changes occur in the middle of musical phrases, so if dither is audible it should be easy for you to hear the switch from dithered to truncated.

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.

I welcome your suggestions for what I could post that will be a valid test, yet not allow people to cheat.

--Ethan

 

[schwaaa]

I'd point out that "artificial" situations (compression of a synthesized sound, for example, then rendering to a lower bit depth) can easily result in audible quantization noise. I wouldn't say that correcting that noise is buying into some sort of brainwashing.

 

[Ethan Winer]

it's a tempest in a teacup nonetheless, compared to a dozen other issues.

I realize you are not arguing too strongly that dither is necessary, but I still have $100 that says you cannot pick it out better than random chance on any musical material using any flavor of dither.

Glen, I'll ask you for the same as I ask others: If dither really can be heard or detected or felt or whatever, please tell me what sort of test I can post that you agree will be valid, to settle this for once and for all.

--Ethan

 

[Ethan Winer]

Also, guys, just to be clear, I'm not trying to be a jerk about this, and I hope nobody takes offense. I made a New Years resolution to work harder to stamp out audio BS, and dither seems as good a place to start as any.

BTW, this is the third forum where I asked a bunch of "dither believers" what sort of test they'll accept as valid, and then the thread suddenly died.

--Ethan

 

[BushmasterM4]

I started a similar thread at gearslutz. Ethan knows (Dithering 101). I did a tune yesterday and used dithering on one edit and none on another. The tune had very low passages as well as loud. I had my son rename them so he knew which was which and I couldnt hear a bit (pun) of difference. So I guessed, and I was wrong.

 

[SouthSIDE Glen]

I realize you are not arguing too strongly that dither is necessary, but I still have $100 that says you cannot pick it out better than random chance on any musical material using any flavor of dither.

I've already agreed with you on that point, Ethan. And I also described why that is a red herring of a bet. In the same way that I'd bet you $100 that I could play you two mixes side by side and not tell which one used a graphic EQ. That does not render the EQ irrelevant. It's a false standard of a test result

Glen, I'll ask you for the same as I ask others: If dither really can be heard or detected or felt or whatever, please tell me what sort of test I can post that you agree will be valid, to settle this for once and for all.

Again, already asked and answered in that it would require a full study with a base of data, and not just one specific test. In detail:

It would require double blind testing of a variety of music content types with a variety of dithering algorithms, with everybody from audiophiles and MEs to college kids and construction workers as the judges/test subjects. This would include control groups with no dithering at all well as all dithering all the time.

Furthermore, such a test would need to be open-ended on it's analysis of the results, with no expectation of a single binary or trinary judgement (e.g. "dithering" is bad", "dithering is good" or "dithering is useless"). Question #1 it should answer is "is dithering ever audible" (followed by percentile breakdown), followed by Question #2 "when audible, is it considered desireable" (followed by percentile breakdown). From there, assuming #1 is answered in the affirmative, further statistical analysis could be performed such as if certain dithering types work better for some people or or some music types, etc.

G.

 

[BushmasterM4]

This is also like the test I was gonna post. Someone said they could hear a difference between a miced guitar amp and a processor unit (Boss GT8, Pod, etc..). So I was gonna post samples of all and see if they could guess. Well never got a reply.

 

[pipelineaudio]

http://shup.com/Shup/17217/dither.zip

something Im working on...could work as a test bed, though reaper's 64 bit path sort of throws it off

 

[masteringhouse]

I'll try to post some info as soon as a get a bit more free time. The effects of truncation versus dither are much more audible at lower levels because the relationship of quantization noise to signal is greater there and more readily audible (correlated). Just as the noise is a piece of gear that has a lower signal to noise ratio can be masked to sound "as good as" a higher end piece of gear, it doesn't mean that they are equal.

Great audio isn't about sweeping issues under the rug, it's about correcting them and making audio the best possible.

 

[mwthompson]

This reminds me of another argument about cables.

Like the difference between Monster and other brands, where scientific test equipment shows differences so minute that a dog would not be able to hear them (between like gauges), but people still swear their guitars sound better with $150 ten foot cables.

I don't have the ears to say they're wrong, but I do favor science over marketing dogma.

This is completely off topic. Sorry.

I have no idea about this dithering issue and this thread is fascinating. I think I will err on the side of doing what everyone else does. (Especially if it's inaudible to most...)

 

[SouthSIDE Glen]

This is also like the test I was gonna post. Someone said they could hear a difference between a miced guitar amp and a processor unit (Boss GT8, Pod, etc..). So I was gonna post samples of all and see if they could guess. Well never got a reply.

Those of us who have been around the forum scene for a while (not to mention real life! ) have seen this kind of thing happen quite often on an informal basis.

There was one "test" done by Farview (I think) a while back where he demonstrated the result of proper miking technique of a git cab and waited for the results to come in. After a multitude of posts from folks praising the sound an trashing the idea of using an emulator, Farview then revealed thr truth that he actually didn't even mic anything; they were indeed listening to a Pod emulation instead.

Another was a conversation snippet where someone (was it NL5? xfinsterx? I don't remember for sure) posted a clip, and someone else (I'll leave this name out...no it wasn't me ) came back and said something along the lines of "It sounds like you got your new analog board up and running finally. I can hear the difference, it sounds sweet." To which the OP replied something like, "umm, sorry, this was all done in the digital box before I got the board hooked up."

My favorites are those occasions (this has happened a couple of different times since I've been here) where someone has posted a clip letting people think on their own that it was an HR clip and asking their opinion as to whether it sounds like a pro job or not. Only after getting a slew of negative replies will the OP come back and let them know the truth that they picked a clip done on by a top band on top gear by a brand name engineer/producer at a world-class studio as soundig definitely amateurish in quality.

Now, I'm sure Ethan's "expectations bias" plays into these situations quite a bit. This is why any such "test" has to be a comprensive study done to accepted standards that reduce all chance of bias in both the test and the judges to statistical zero.

But it also shows, perhaps even more importantly, that perception is everything. Unfortunately.
G.

 

[Ethan Winer]

I'd bet you $100 that I could play you two mixes side by side and not tell which one used a graphic EQ.

Now hold on a minute. Are you saying I couldn't hear that the files are different? Or just that I can't identify which one was the original and which was the EQ'd version? If a track contains frequencies that align with what is EQ'd, and the EQ is at least 1 dB with a reasonable Q (2 or lower), I'm sure I can tell that the files are different.

And this is exactly my point. I contend that nobody can tell that the dithered versus truncated files are even different at all. If I play you a random mix of the files - not just A B A B but maybe A A B B B A A - you won't even be able to group the A and B files together.

It would require double blind testing of a variety of music content types with a variety of dithering algorithms, with everybody from audiophiles and MEs to college kids and construction workers as the judges/test subjects. This would include control groups with no dithering at all well as all dithering all the time.

LOL, this brings up another great point. From my perspective, the only time a blind test is needed is to "prove a negative" so to speak. That is, you have a bunch of believers who claim to distinguish dither, jitter, 192 KHz sample rate, etc. They will not accept that they can't hear it, so in that case a DBT is needed to prove it to them. But for things that really do matter, a blind test is never needed because everyone can hear it. For example, we do not need a DBT to distinguish 8 bits versus 16, or 22.5 KHz sample rate versus 44.1 because the degradation is obvious.

So if dither really matters, and is audible, then it should be audible to all every time.

--Ethan

 

[Ethan Winer]

I don't have the ears to say they're wrong, but I do favor science over marketing dogma. This is completely off topic. Sorry.

You do have the ears, it's not off-topic, and you have no reason to be sorry.

Seriously, I see dither as almost the same as speaker wire. The only difference is speaker wires are not even a little different (unless they're incompetent, which happens), where dither does change the sound but just not enough for anyone to hear it.

--Ethan

 

[mattkw80]

This reminds me of another argument about cables.

Like the difference between Monster and other brands, where scientific test equipment shows differences so minute that a dog would not be able to hear them (between like gauges), but people still swear their guitars sound better with $150 ten foot cables.

I don't have the ears to say they're wrong, but I do favor science over marketing dogma.

This is completely off topic. Sorry.

I have no idea about this dithering issue and this thread is fascinating. I think I will err on the side of doing what everyone else does. (Especially if it's inaudible to most...)

+1 on all that.

Including the bit about cables.

It also makes me wonder what else is "BS baffles Brains."

Ethan - I think you new years resolution is great.

I also can't wait to see the results of your challenge, and as well, the rest of us will greatly benefit from the other examples being put together by the opponents of your theroy.

A week from now, I can't help wonder if the agree'd upon answer is........... "dither matters, sometimes."

 

[pipelineaudio]

So if dither really matters, and is audible, then it should be audible to all every time.

--Ethan

I think you should make that a bit more specific. I don't think anyone is saying it should ALWAYS be audible, just at extremely low levels.

I'm of the opinion that we can truncate most cd's made in the past ten years to 8 bits, and except for a fade out at the end, for the most part it will sound the same.

It seems to me though, if I REALLY push some gains around at extreme settings that dither MIGHT make an audible difference with your speakers up all the way listening to a ridiculously low level part.

 

[Ethan Winer]

if I REALLY push some gains around at extreme settings that dither MIGHT make an audible difference with your speakers up all the way listening to a ridiculously low level part.

Yes, if you have low level content and raise the volume unnaturally, such that normal passages on the same track would blow out your speakers, I agree that dither might be audible. This also requires a very low ambient noise floor on the recording. But if you play a track as loud as one could reasonably tolerate, and then listen as a reverb tail fades out, and you don't raise the volume further, you won't be able to tell.

--Ethan

 

[SouthSIDE Glen]

Now hold on a minute. Are you saying I couldn't hear that the files are different? Or just that I can't identify which one was the original and which was the EQ'd version?

That you couldn't tell which one was which. Additionally - sorry, a whole new angle here - that you'd not necessarily like the same one tomorrow that you like today.

If I play you a random mix of the files - not just A B A B but maybe A A B B B A A - you won't even be able to group the A and B files together.

That could very well be true. But is that testimony to some kind of psychological effect on my part or to the subtlety of the effect? The result in and of itself does not yield a definitive answer. And conversely, if you say you can't hear a difference, how do you or I know that means there is no differece, that you're ears can't perceive a differece (this time), or that your own negative bias is affecting you the same way you say a positive bais is affecting me?

All of this is why

the only time a blind test is needed is to "prove a negative" so to speak.

is IMHO an incorrect perspective to take. There are REASONS that tests need to be conducted with rigor; they need to account for all variable and eliminate any ambiguity of result. One of the main reasons whey we keep hearing contradictory news on TV about all these different heath studies where one says "eating chocolate is bad for your health" and another says "eating chocolate is good for you" is because the methodology of the studies is glossed over and, in many cases, is quite faulty. It doesn't matter what the objective of any given test is; if not desigend right, it will give false results.

So if dither really matters, and is audible, then it should be audible to all every time.

Why? That reasoning does not follow. Here's an analogy:

Weather is known to affect people's health (humidity affecting atrthitis, barometric pressure affecting sinuses, etc.) Many are so in tune that the can predict a storm coming by the "feeling in their bones". Though this is not 100% accurate, necessarily, there can be other causes than an incoming storm for that feeling. Many people go years or their entire lifetime without making the correlation between weather and their body, others suspect something is going on there but just can't pick out a pattern. Some are affected by the same weather changes in entrirely different ways. Others are not affected by weather changes at all.

Does this lack of ubiquity make the effect irrelevant or non-existant? No, it just makes it more subtle and complex than many of us are comfortable in accepting. But it reamins real nonetheless, and not just old folk tales.

Dither is, at best, a very subtle effect. Not everybody will always hear it. Not every that can hear it will always ID it, or necessarily even hear it every time. It's effect - or perhaps lack thereof, to take in your position as well - is not as easily qualifiable as it is quantifiable. Many real effects in life are like that. That does not make them unreal, it only makes them esoteric.

Personally, I think that if dither were to disappear from the universe tomorrow, that very few would even notice, let alone care. My Coleman Hawkins recordings will still send me there whether they have been dithered one way or another or not. I also do not buy into Monster Cable or 96k recording, just for the record; any positive benefits of either are usually attributable to some other factor than the claims actually in question.

But I do see dithering as something more than a waste of time; while the effects may not slap anyone in the face, nor even be universal in their subtlty, I do think, based upon my own experiences that there is *something* subtle and too complexly patterened there to disregard in the same psycho-mythical basket as most other audiophilia.

I just don't bother losing any sleep over it when that ditering is at the end of the crapola signal chain I have .

G.

 

[Ethan Winer]

It also makes me wonder what else is "BS baffles Brains."

Jitter is BS because at -100 dB or more it's even softer than dither. Ultra-high sample rates are BS because nobody can hear past 22 KHz. Phase shift is BS because it's inaudible. I'm sure I can come up with more!

--Ethan

 

[Ethan Winer]

That you couldn't tell which one was which.

In that case of course I agree because you could take a thin track and make it bassy, or it could have been bassy and you EQ'd it to be thin. But I could tell they are different!

Additionally - sorry, a whole new angle here - that you'd not necessarily like the same one tomorrow that you like today.

Agreed completely. Human hearing is very frail and subject to change at a moment's notice. Give me a coupla hits of hash and everything I play will sound much better.

But I'm not talking about psychological effects which is a whole 'nother ballgame.

There are REASONS that tests need to be conducted with rigor; they need to account for all variable and eliminate any ambiguity of result.

I agree, and you made a good point with your example "Chocolate is bad for you. No wait, it's good for you. No, sorry, it's bad." But that is a far more complex issue for many reasons. It takes years and years to track thousands of people, all with different physiologies and varying health due to other issues. This is a much simpler situation - either it can sometimes be detected reliably or it can never be detected reliably. I say the latter.

Dither is, at best, a very subtle effect.

Tell that to NL5 here ("I can tell right away"):

http://www.homerecording.com/bbs/showpost.php?p=2848016&postcount=10

And pezking here ("horrific"):

http://www.homerecording.com/bbs/showpost.php?p=2848282&postcount=21

I also do not buy into Monster Cable or 96k recording, just for the record; any positive benefits of either are usually attributable to some other factor than the claims actually in question.

If you extend "usually attributable to some other factor" to dither we'll be in full agreement!

Good discussion guys. As soon as we can agree on a test method, I'm confident you'll all side with me in the end.

--Ethan

 

[SouthSIDE Glen]

Agreed completely. Human hearing is very frail and subject to change at a moment's notice. Give me a coupla hits of hash and everything I play will sound much better.

Hash? Where are you, Amsterdam???? Damn, I haven't even *seen* hash around my parts for 20 years . But you're right. A little weed can go a long way to making dither sound like Luciano Pavoritti.

But I'm not talking about psychological effects which is a whole 'nother ballgame.

Agreed. But I do make the point that the subtlety is such where it's difficult to seperate the two. All the more reason for a REAL testing procedure.

If you extend "usually attributable to some other factor" to dither we'll be in full agreement!

I have no data, but I hold dither above the others because IME there does seem to be a "signal" there (speaking metaphorically) that rises above the background noise (again metaphorical) that IME does not readily fit into a bias or phycho- effect.

Good discussion guys. As soon as we can agree on a test method, I'm confident you'll all side with me in the end.

Agreed on the discussion . I look forward to more like this.

I've put a test method on the table that is quite doable over the Internet. In fact, I'd be happy to put it together and host it over at IRN (after I get some other development work done there that is currently on the operating table, that is.) Everybody who has mentiond here they have samples they want to use, I'll take all comers, using as many examples as make sense. Anybody interested in contributing, can contact me via the e-mail link on this BBS.

I will remain open, though, as to the results. I will side with you IF the results tell me to as long as you can say the same for me .

G.