Dither - My Ears can't hear it

[Ethan Winer]

Neither of your files resembles the original sine wave, even though the recorded level is at a very reasonable -14 dB.

I just read your page again and saw that you actually did the processing with the signal 14 dB lower than in the 24 bit file on your site. So I took your 24-bit file, dropped the volume by 14 dB, converted to 16 bits truncated, then raised it 14 dB.

I still get a sine wave that looks and sounds like the original. So even though I mis-read your web page the first time, it's clear you reduced the file to 8 bits instead of 16 or something like that.

--Ethan

 

[masteringhouse]

Ethan -

The description of what I did to my files is explained on the page, there were no errors, just misunderstanding on your end. The files were originally recorded at -80 dBFS and then brought up to -14 so that they could be heard more easily and without additional noise from a montoring chain. Truncation distortion is correlated to the signal, that basically means that at lower levels the distortion is a larger percentage of the signal and can be heard readily.

Even on your graphs you can start to see where the truncated signal gets "squared off". This is essentaily harmonic distortion, something to be avoided, not swept under the rug of a louder signal.

You're right about one thing though, I do need to get out more.

So who are these other engineers who claim truncation distortion and jitter are a farse used to sell expensive audio gear? Certainly not Bob Katz who has several pages on the subject (or is he trying to sell books?)

Ethan, please don't take my comments personally, you seem like a really nice guy. But when someone starts coming in with guns blazing trying to dismiss well-founded digital theory as "myths", well I gotta wonder what their intentions really are, or if they are just misinformed.

 

[masteringhouse]

I just read your page again and saw that you actually did the processing with the signal 14 dB lower than in the 24 bit file on your site. So I took your 24-bit file, dropped the volume by 14 dB, converted to 16 bits truncated, then raised it 14 dB.

I still get a sine wave that looks and sounds like the original. So even though I mis-read your web page the first time, it's clear you reduced the file to 8 bits instead of 16 or something like that.

--Ethan

Yep, they were not reduced to 8 bits though. They were reduced to 16 bits within the 24 bit bus of a Pro Tools HD system (non-dithered mixer) using an L2 plug-in. If you try to increase the gain after a conversion to 16 bit, well it gets real ugly.

One thing to possibly consider is what Sound Forge is doing "behind the scenes". I'm not aquanted with this program at that level, but it could be doing it's processing at 48 bit. Also try using the FFT instead of viewing the waveform.

 

[CIRO]

Hey, "MASTERING" ...
HAPPY BIRTHDAY,TOM!!!!!!!!!!!!!!!!!!!!!!!!

Ciro
www.soundclick.com/openstation

 

[masteringhouse]

Ethan -

I just tried the same procedure in Sound Forge only using it's TPDF dither. Pretty much the same result as my Pro Tools experiment:

16 bit TPDF dither:
http://http://www.masteringhouse.com/dither/SF_16_bit_TPDF_dither.jpg

16 bit truncated (note additional harmonic distortion present):
http://www.masteringhouse.com/dither/SF_16_bit_trunc.jpg

 

[masteringhouse]

Hey, "MASTERING" ...
HAPPY BIRTHDAY,TOM!!!!!!!!!!!!!!!!!!!!!!!!

Ciro
www.soundclick.com/openstation

Thanks Ciro!

Shows what an audio geek I am, sitting here playing around with dither on my B-Day.

I hear that Ethan is going to be getting me that Weiss unit for a B-Day present

 

[Ethan Winer]

Tom,

First, screw dither, happy birthday.

Okay, break's over, back to the dither.

The files were originally recorded at -80 dBFS and then brought up to -14

You're kidding, right? This is what you offer up as proof of the importance of dither? Okay, you're correct, it's not 8 bits - it's more like 3 bits. Please tell me this is a joke.

Ethan, please don't take my comments personally, you seem like a really nice guy.

Back atcha.

when someone starts coming in with guns blazing trying to dismiss well-founded digital theory as "myths", well I gotta wonder what their intentions really are, or if they are just misinformed.

How am I misinformed? Dither is indeed inaudible and pointless for normal music recorded at sensible levels. That you had to resort to 3 bits to make your case proves my point, not yours.

I hear that Ethan is going to be getting me that Weiss unit for a B-Day present

I shipped it FedEx overnight so you'll have it tomorrow. Happy birthday!

Tom, as soon as you or anyone else posts an example proving that dither actually matters, I'll change my opinion in a New York minute. Until then, all the theory in the world is meaningless. Also, I never said that dither is useless in all situations or that the theory is not sound. Some things, like low-bit-rate telephony, are indeed improved by dither. And if hardware samplers still used only 12 bits, they too might benefit from dither. But in current practice it is a pointless exercise when going from 24 bits to 16, and fretting over stuff like dither and jitter only distracts folks from what really matters.

--Ethan

 

[masteringhouse]

You're kidding, right? This is what you offer up as proof of the importance of dither? Okay, you're correct, it's not 8 bits - it's more like 3 bits. Please tell me this is a joke.

Only to make it obviously audible Ethan. For some this is audible given nearly any level, for some even the type of noise shaping is audible. You have to know what to listen for to start with in order to hear it. It's the same way with compression to some starting off in audio, they don't hear it until they have worked with a compressor for a while or until you exaggerate its effect.

It's all subjective I suppose, like most things audio, and I guess no amount of debate is going to convince either of us to give up on our points of view. I do agree that there are more important issues to be concerned with, like the actual signal above dither! But likewise I feel every little bit (no pun intended) is important and shouldn't be ignored. Especially when it's adding a nasty component that is so easily gotten rid of.

Ah well, Rock on Ethan!

 

[Ethan Winer]

For some this is audible given nearly any level, for some even the type of noise shaping is audible. You have to know what to listen for to start with in order to hear it.

I know what to listen for and I can't hear it at all. Ever. On any material. This is why I keep begging for someone to post a file we can all listen to and say, "Aha, so that's the difference between dithered and truncated."

I'm still waiting, and I imagine I'll still be waiting on my deathbed.

Ah well, Rock on Ethan!

Indeed.

--Ethan

 

[The New Hype]

the question is not being able to hear or not to hear the dither noise. the question is how it affects the perception of the whole sound. dithered sounds smoother and truncated sounds granier. it's extremely subtle and i don't think anyone could point where is it switched on and off in ethan's example, cos there are some musical changes that can be confused with dither stepping in or out. but there IS a minor improvement when using dither that i would never discard. c'mon, it's so easy to put it in. why not?

 

[mattkw80]

the question is not being able to hear or not to hear the dither noise. the question is how it affects the perception of the whole sound.

The perception of the whole sound ?

You perceive the sound with your ears...... and your ears can't hear dither.

Not sure why you would say "the question is not being able to hear or not to hear the dither".

Sounds like your ready to put the $100 up for Ethan's challenge.

Not sure how you'll fare however, considering the thing you'll be listening for can not be heard.

 

[The New Hype]

i meant the loss of fidelity prevented by the dither can be HEARD (and this is with the ears, believe me), if you compare a truncated song with the same one dithered, if you a/b the same parts... i'm not up for ethans test and in my last post i said that i believe no one in the world is (sure i might be wrong, but well...). that stuff is not the same sound dithered vs. truncated, the musical information changes every second there, so it's impossible to know which sector is truncated and which is not, cause the changes caused by dither are far more subtle than the changes caused by the flow of the music itself. but if you say there's no real change with dither at normal levels, nothing audible at all, i strongly disagree.

 

[MCI2424]

i meant the loss of fidelity prevented by the dither can be HEARD (and this is with the ears, believe me), if you compare a truncated song with the same one dithered, if you a/b the same parts... i'm not up for ethans test and in my last post i said that i believe no one in the world is (sure i might be wrong, but well...). that stuff is not the same sound dithered vs. truncated, the musical information changes every second there, so it's impossible to know which sector is truncated and which is not, cause the changes caused by dither are far more subtle than the changes caused by the flow of the music itself. but if you say there's no real change with dither at normal levels, nothing audible at all, i hardly disagree.

Perhaps the "change" that can be heard has more to do with the fact that all the data went through a unique algorithm and the 1 thing that algorithm was meant to change is inaudable? There are a few ways to write the software to do this, and if they all sound different, then it probably is not "dither' you are hearing at all. If dither was the result, all dither programs would have the exact same affect on the data.

 

[masteringhouse]

Perhaps the "change" that can be heard has more to do with the fact that all the data went through a unique algorithm and the 1 thing that algorithm was meant to change is inaudable? There are a few ways to write the software to do this, and if they all sound different, then it probably is not "dither' you are hearing at all. If dither was the result, all dither programs would have the exact same affect on the data.

Good theory, however one question is if one can you hear a difference between a 24 bit file and the same one that has been truncated and brought back to 24 bit?

If no, then we should all be recording, mixing, and mastering at 16 bit to save a lot of bandwidth and disk space.

"I love the 80's" (not)

 

[Ethan Winer]

one question is if one can you hear a difference between a 24 bit file and the same one that has been truncated and brought back to 24 bit?

Actually, I have a test of exactly that on my web site here:

http://www.ethanwiner.com/BitsTest.html

Not only dithered and truncated to 16 bits, but also truncated to 13, 11, and even 9 bits.

This page and files have been on my site since 2002, and so far I've received 66 guesses as to which files are which. I say "guesses" because that's exactly what they've all been. Of the 66 emails I've received, two people nailed it and identified all five files correctly. But hold on a moment - don't get too excited! This is slightly less than chance from random guesses. Random guessing should yield 1 in 25 correct (5^2), so I'd expect 2 correct for every 50 entrants. Which is about what I've received.

--Ethan

 

[NL5]

Actually, I have a test of exactly that on my web site here:

http://www.ethanwiner.com/BitsTest.html

Not only dithered and truncated to 16 bits, but also truncated to 13, 11, and even 9 bits.

This page and files have been on my site since 2002, and so far I've received 66 guesses as to which files are which. I say "guesses" because that's exactly what they've all been. Of the 66 emails I've received, two people nailed it and identified all five files correctly. But hold on a moment - don't get too excited! This is slightly less than chance from random guesses. Random guessing should yield 1 in 25 correct (5^2), so I'd expect 2 correct for every 50 entrants. Which is about what I've received.

--Ethan

I emailed my answers and never heard back....... it seemed pretty easy, other than two of the files were really close.

edit - found my answer -

I have them as 5. 3, 4, 2, 1 although I am unsure of 3 and 4 it was a tossup. (worst-best)

 

[SouthSIDE Glen]

This page and files have been on my site since 2002, and so far I've received 66 guesses as to which files are which.

Ethan, I don't know how many more different ways there are to try to explain to you that that test is meaningless as it is set up. It is bad science of the type that fills the "paranormal" and "intelligent design" bookshelves of your local Borders.

I don't think I have heard anyone in this thread say that they could *identify* dither in a sound test. "Dither" does not have a sound of it's own, and to set up a test asking for that is just setting people to to take an almost guaranteed fall. All that test proves is something that all but the most hyperbolic of pro-dither engineers already stipulate.

Second, you keep referring to "dither" as if it's a single entity. There are many different dither algorithms of varios quality and of various color of effect on different recordings of different nature. To not take that into accont and to treat "dither" as a single entity is like referring to "the common cold" and asking why there is no cure. (Hint: because there are several hundred different causes; there is no *one* "common cold".)

All in all, though, I do wish this thread would die rather than keep spiraling around in circles ad nauseum. The fact that dither's effect is measurable always, is audible sometimes by some people (whether it's psychoacoustic or consciously audible is irrelevant), and costs virtually nothing in time or effort to execute shuld be enough to put the matter to rest right there.

G.

 

[The New Hype]

This is slightly less than chance from random guesses. Random guessing should yield 1 in 25 correct (5^2), so I'd expect 2 correct for every 50 entrants. Which is about what I've received.

ethan, this is not definitely relevant, but in fact, 2 in 66 is above the chances from random answers. the correct formula for this is 5x4x3x2 = 1 in 120, or 2 correct answers for each 240 tries.

i heard the samples and i think it's not really easy to point which is which with no error. but there are differences between them, this is obvious. so, what are you trying to prove? some people, in a blind test, can actually prefer a lossy format sound, as mp3, besides a 24 bit crystal-sounding pcm. psychoacustic madness, you know. but they ARE different, we all know they are. if what you are trying to say is "11 bit and 24 sounds the same"... lol... no, man, they do not.

AND... at lower bitrates, the signal degradation introduced by any DSP is more apparent. this can be measured and it's not good, i really don't think it's a good idea to mix, record and etcetera at a lower bitrate. it's hard to believe you are serious about it.

i really respect your skepticism, but man.............. no.

(and sorry for my poor english. cheers)

 

[SouthSIDE Glen]

ethan, this is not definitely relevant, but in fact, 2 in 66 is above the chances from random answers. the correct formula for this is 5x4x3x2 = 1 in 120, or 2 correct answers for each 240 tries.

That's a good catch, Hype, I missed that. You're right, the odds are calculated as 5! (the factorial of 5) and not 5^2. While 66 attempts may not be enough of a database to be statistically solid, it is so far showing a success rate some 3.6 times greater than a random selection in favor of dither!

Of course, I still stand by the position that the test is ultimately extremely flawed for the reasons given before. But the irony is that even if I'm wrong about that, and the results stayed fairly close to what they are now, the test would actually support the proposition that dither is indeed statistically signifigant.

G.

 

[NL5]

That's a good catch, Hype, I missed that. You're right, the odds are calculated as 5! (the factorial of 5) and not 5^2. While 66 attempts may not be enough of a database to be statistically solid, it is so far showing a success rate some 3.6 times greater than a random selection in favor of dither.

Of course, I still stand by the position that the test is ultimately extremely flawed for the reasons given before. But the irony is that even if I'm wrong about that, and the results stayed fairly close to what they are now, the test would actually support the proposition that dither is indeed statistically signifigant.

G.

The other problem being missed here is that the majority of people here are listening in a less than ideal environment, with less than ideal equipment. I have never done a dither test other than on my own mixes, and Ethans, but I have done several converter tests, and have gotten every one right. Despite what Ethan has said, there is a difference in converters. If you are trying to tell which is a Lavry and which is a firepod on a m-audio or soundblaster thru a pair of rockit-5's in a bedroom with little or no treatment, you won't ever be able to tell.

Ethan's bit test seemed pretty obvious as well (other than the two being pretty close - both the 11 and 13 bit files sounded equally bad.) I'm curious to see if I was right.

Last, I will agree with Ethan that dither is one of the last things most people need to worry about. Heck, 24 or 16 bit won't make much of a difference in most recordings. there are much bigger fish to fry. However, once you start to get things coming together, every improvement is subtle. Very subtle. It's getting all those subtleties to add up that separates the pro's from the rest of us.