Dithering 101

[bob katz]

Originally Posted by Ethan Winer

I don't see why this would be so. Let's take the very obvious test case of rolling off everything sharply below 200 Hz. Anyone can tell that blind or not blind, right?

So if I put together a track and switched in a 200 Hz low-cut filter at various times, 100 people out of 100 will know where I did that, and which was flat and which was rolled off, yes?

You would be SURPRISED to discover that these differences are not so obvious at all when it is done as a switch within a continuous piece of music! Been there, done that.

Here's a little story:

One day 40 years ago, in my bachelorhood I lived with a bunch of roommantes. My roommate was in the next room listening to music on a pair of headphones. I was in the next room where the ampliier was, and just to be mean and nasty, I very slowly moved the stereo balance control a little bit towards the right, a tiny bit, and increased this every 5 minutes. By the time a half hour had passed, most of the sound was only coming into his right ear. At which point he finally got wise to this, jumped out of his seat and jumped to the stereo control. I was just smirking in the next room.

The point is that we get accustomed to certain sounds. Listen to an old movie on TV and the high pass at 200 Hz is not so obvious at first, unless we are very trained listeners. The bottom line is, even for your high pass test, without access to the original material, we may not notice or know that it has happened. ESPECIALLY if we are not told exactly where the switch occurred. Because in fluid music, there is not always low frequency information. And in addition, psychoacoustically, the ear/brain supplies missing fundamentals in a low cut situation!

So it is not as cut and dried as you propose. Your test method is fatally flawed. See my other post today about cheating and the correct way to conduct such a test.

[johnlink]

Originally Posted by bob katz

My roommate was in the next room listening to music on a pair of headphones. I was in the next room where the ampliier was, and just to be mean and nasty, I very slowly moved the stereo balance control a little bit towards the right, a tiny bit, and increased this every 5 minutes. By the time a half hour had passed, most of the sound was only coming into his right ear. At which point he finally got wise to this, jumped out of his seat and jumped to the stereo control. I was just smirking in the next room.

Bob, I hope that you've included this story in your bio for the second edition of your book, as it reveals a great deal about you!

John Link

[twwalsh]

Originally Posted by Ethan Winer

Here's where got the idea, plus many similar posts over at Lynn's forum:

TWWALSH - i can hear the 16 bit dither in voxengo elephant IMMEDIATELY and unmistakably

stating that i can hear dither unmistakably and claiming that it's critically important to a master are two very different things. i was stating the former.

[rentedroom]

Thank you Ethan for your challenge about dither.

I have done my own tests with dither and there might be applications where its valid, but I notice that all of these units with dither actually adds a negative coloration to the final master that has not been discussed in this thread.

The units I have tested are:

1. Weiss DS1-mk2 - Pow-r 1, 2 & 3.
2. Cranesong head dither. In the digital mode.
3. Waves L2 - Hardware dither.

The fact that there is such s heated discussion about dither makes me wonder if anyone is listening to what these processors are sounding like over what the dither is actually doing.

[peeder]

I think if you guys take a cursory look at my spectrograms posted earlier you'll see where dither matters and where it doesn't.

Your test is going to depend very vitally on the source material. If the material is open and not particularly loud and has tails decaying down below -50dbFS frequently, you'll be able to pick out the dither in a quiet and full-spectrum listening environment handily. If the material has a closed arrangement and is loudness maximized you'll have a much harder time...you might be best off listening to the very start or very end of the file over and over seeing if you can pinpoint the difference. But if you do the within-file switching thing the start and ends might not even be different in different test cases.

If you want to do a test where you have a series of test subjects in which every ABX is increased 10db in level from -140dbFS to full scale, you will see the point at which the level drowns out the dither in practice for that particular test material (I would predict at about -40dbRMS for dense material, although you might be able to hear the noise floor differences still at that level depending on how closely the sample was cropped).

[bob katz]

Originally Posted by rentedroom

Thank you Ethan for your challenge about dither.

I have done my own tests with dither and there might be applications where its valid, but I notice that all of these units with dither actually adds a negative coloration to the final master that has not been discussed in this thread.

The units I have tested are:

1. Weiss DS1-mk2 - Pow-r 1, 2 & 3.
2. Cranesong head dither. In the digital mode.
3. Waves L2 - Hardware dither.

The fact that there is such s heated discussion about dither makes me wonder if anyone is listening to what these processors are sounding like over what the dither is actually doing.

Dear rented room:

You seem to be throwing a misinformed monkey wrench in these works!

Let me answer your comment with some simple facts:

1-- There is USUALLY a subtle sonic difference when dithering from 24 to 16. Perhaps Ethan hears that, perhaps not, but we can all agree it is very subtle, especially to an untrained ear. There is often a slight loss (loss of separation, depth, warmth) compared to the original 24 bit. HOWEVER, truncation is ALWAYS objectively worse than dithering. What you are advocating (not dithering) is simply not acceptable once you recognize the distortion and more severe losses due to truncation. If you prefer to truncate and hear the distortion, aliasing, cold sound, and losses of depth, then that's your prerogative, but it is not the proper choice for any critical listener.

2---Regarding the Cranesong. You are mistaken: IN digital mode it DOES NOT DITHER AT ALL! There is NO dither in the Cranesong HEDD when run D-D.

3---The dither in the Waves L2 is another form of noise shaped dither, different in shape than PWR, and is another valid way to get from 24 to 16.

Whatever degradation or "big losses" you are claiming to hear are probably FAR less than you imagine and furthermore, not dithering is far more of a loss! (At 16 bits).

Furthermore, you seem to have missed Ethan's point entirely. On the one hand you are cheering him on, while on the other, Ethan's point is that dither is quite subtle in its audibility, while you seem to be on the rampage about all the severe "damage" that dither is doing. You have done nothing constructive to help this thread along, in my opinion...

BK

[Ben B]

Originally Posted by bob katz

Whatever degradation or "big losses" you are claiming to hear are probably FAR less than you imagine and furthermore, not dithering is far more of a loss! (At 16 bits).

Furthermore, you seem to have missed Ethan's point entirely. On the one hand you are cheering him on, while on the other, Ethan's point is that dither is quite subtle in its audibility, while you seem to be on the rampage about all the severe "damage" that dither is doing. You have done nothing constructive to help this thread along, in my opinion...

BK

Misinformed or not, his post did not come across as a rampage to me. Did he ever say that dither does severe damage to audio? I read this as him having objections to the sounds of those units, not necessarily attributing the sound to dither in a generic sense.

Ben B

[jayfrigo]

Originally Posted by rentedroom

but I notice that all of these units with dither actually adds a negative coloration to the final master that has not been discussed in this thread.

Actually I pointed this out a page or two ago, that some people find coloration in noise shaped dithers as opposed to flat TPDF. Near-Nyquist ones like Apogee's UV22 and one of the POW-r types in particular are sometimes reported as having a brightening effect.

Bob's point is a good one though; is that it can't be heard, or that it's can be heard too much and you don't like it? Those are two very different positions. Truncation sounds worst to me, when it sounds like anything at all. My opinion is to use flat dither in many cases, and noise shapers on a case by case basis dependent on program material when you have a reason to object to the flat.

[rentedroom]

I'm not misinformed, the designer and owner of Cranesong, David Hill just confirmed it to me himself that in digital mode you can dither. If you look at the picture (attached to this post) of the Cranesong Hedd you will see in digital mode that you can switch it to 16bit digital. Unless you know something that David Hill doesn't know about his own products. He is the person I got the information from.


As to my post, for clarification, I was stating that using any device , Plug-in, Hardware, etc. with dither, there is different coloration to each process. I was interested to see if that was important to any one else and if there was any other opinions about that.

These expensive devices change the sound, sometimes for the worse.

[Ethan Winer]

Originally Posted by bob katz

The problem is it is probably impossible to avoid cheating.

Then I have no further interest because if people can cheat, then they will just to "prove" their preconceived beliefs. So that means getting a group of people here to my studio, or down to yours in Florida. I won't rule out a trip down to your place if we can truly make a "Gearslutz event" out of it. But first, maybe we can get some folks local to me to my place for the first round of tests? Again, TW Walsh says "i can hear dither unmistakably" so this should not require someone with a PhD in statistics to design a test to settle this for once and for all. Either folks can hear it or they can't.

I'll mention that the last time something like this came up, a mix engineer who frequently posts at Gearslutz agreed to let me test his ability to tell if a 20 KHz low pass filter was inserted into his monitor chain. But then when I tried to arrange a time he never replied further.

I see a few posters in this thread are within a few hours of me in Western Connecticut, so how about it guys? Anyone here care to step up to the plate and see if you can tell when a file is dithered versus when it's truncated? I have two different decent size rooms and can accommodate up to a dozen people or so. Anyone is welcome to supply additional test material if they'd like.

--Ethan

[twwalsh]

i might be willing to come out from boston for a test.

however, ethan - just because i claim i can hear dither well in situations under my control, this doesn't mean i'm claiming to hear it under any circumstances anywhere. that's where bob's criticism of your test comes in. without some kind of parameters, your test is like trying to find a needle in a haystack, and it's VERY skewed towards proving your point. that's why we need a control (undithered 24 bit), and _fair_ presentation of the experimental results (dither vs truncation). switching the dither and truncation and random points throughout a piece of music which is changing constantly in other ways is just an outlandish, unscientific "test".

this is not a pissing contest or an ego thing. dither is subtle. i think that's your point...but it can become less and less subtle to the trained ear. that's my point. sometimes dither helps material, sometimes maybe it doesn't. maybe we should just leave it at that!

[bob katz]

Originally Posted by Ben B

Misinformed or not, his post did not come across as a rampage to me. Did he ever say that dither does severe damage to audio? I read this as him having objections to the sounds of those units, not necessarily attributing the sound to dither in a generic sense.

Ben B

I apologize if I misinterpreted his post! But it seemed to me that he was objecting to dithering per se over simply truncating.

BK

[bob katz]

Originally Posted by rentedroom

I'm not misinformed, the designer and owner of Cranesong, David Hill just confirmed it to me himself that in digital mode you can dither. If you look at the picture (attached to this post) of the Cranesong Hedd you will see in digital mode that you can switch it to 16bit digital. Unless you know something that David Hill doesn't know about his own products. He is the person I got the information from.


As to my post, for clarification, I was stating that using any device , Plug-in, Hardware, etc. with dither, there is different coloration to each process. I was interested to see if that was important to any one else and if there was any other opinions about that.

These expensive devices change the sound, sometimes for the worse.

Switching to 16 bit digital doesn't necessarily mean it is dithered! Possibly I mispoke a mistake (I never make mistakes :-).

I NEVER use anything but the 24 bit position in the Cranesong, preferring to dither elsewhere. I'll make some tests of the 20 and 16 bit modes of the Hedd and get back to you. I am pretty certain that this box just plain truncates in the digital mode.

BK

[Kees de Visser]

Originally Posted by jayfrigo

My opinion is to use flat dither in many cases, and noise shapers on a case by case basis dependent on program material when you have a reason to object to the flat.

I second that opinion wholeheartedly ! Flat dither never sounds obtrusive to me and AFAIK I can't hear it under normal playback conditions. I see no reason not to dither, especially since it comes at no extra cost.

Something else I've noticed: the golden rule that dither should be applied only once has become so widespread that many users prefer to truncate several times and only dither the last step because that's what they've been told to do. It doesn't seem a good idea to me, but Ethan's dither test might give some food for thought.

Ethan, additional (dither) noise can(!) have (un-)masking effects on the music that won't be present in the 24-bit original. How about creating an additional 24-bit testfile consisting of the 24-bit original with added 16-bit level (24-bit resolution) dither to it, to avoid noise differences with the 16-bit dithered version?

[bob katz]

Originally Posted by bob katz

Switching to 16 bit digital doesn't necessarily mean it is dithered! Possibly I mispoke a mistake (I never make mistakes :-).

I NEVER use anything but the 24 bit position in the Cranesong, preferring to dither elsewhere. I'll make some tests of the 20 and 16 bit modes of the Hedd and get back to you. I am pretty certain that this box just plain truncates in the digital mode.

BK

Having gone on memory from tests I did years ago when I first got the HEDD-192 I did further testing. Here goes:

DO NOT EVER USE THE CRANESONG HEDD IN DIGITAL MODE EXCEPT IN THE 24 BIT POSITION. If used in this way and fed a 24 bit source signal and there is no DSP going on (bypass mode), then it will produce a bit-transparent clone of the source. However, the other modes (digital 16 and 20) are not adding dither, there is always output down to the 24th bit in all three positions when set to digital and possibly some stuck LSBs bits in the 16th and 20th position so I would not use those. For the record, I recommend the 24 bit position of the HEDD at all times in all modes, and using external processors to perform any wordlength reduction. Basically, dithering properly and sufficiently randomly is not the HEDD's strong point or intended purpose.

As far as the poster's lumping the HEDD, Weiss, Cranesong and L2 together and not liking the sound of ANY of them, then I wish you luck. Because between PWR 1, 2, 3, and the L2s shapes, including flat and shaped dither, you have a large enough assortment of dither flavors all properly implemented, by the way, to satisfy anyone at all! So I don't know what your complaint is, but I do suggest some more critical listening, examination, and testing on your part.

BK

[jayfrigo]

Originally Posted by bob katz

So I don't know what your complaint is, but I do suggest some more critical listening, examination, and testing on your part.

BK

I suggest the less aggressive shapers or flat TPDF as a good place to start, and see where it goes from there. I engaged the L2's "U" or ultra position once...

Once.

[Ethan Winer]

Originally Posted by twwalsh

i might be willing to come out from boston for a test.

That would be awesome! Drinks afterward are definitely on me.

that's why we need a control (undithered 24 bit), and _fair_ presentation of the experimental results (dither vs truncation).

That's why I'm glad to have you bring source material. We can truncate and dither together in Sound Forge to prepare the test from your files, and the only time I'll ask you to look away is when I rename the files prior to playing them.

this is not a pissing contest or an ego thing.

Absolutely! This is pure science, to determine for once and for all whether dither is audible under even the most favorable circumstances.

dither is subtle. i think that's your point

Take that one step further to "dither is never audible, period" and you understand my point. Of course, this is when reducing 24 bits to 16 for "normal" program material. Not 8 bits to 4 bits, or with music so soft it peaks at -30 dB, etc.

Again, I am willing to be proven wrong. And I'll be the first to admit it publicly, and post in every thread from here forward that dither is audible. But first it has to be proven to my satisfaction.

It would also be great if a few more people would attend. Anyone?

--Ethan

[Lupo]

Originally Posted by Ethan Winer

We can truncate and dither together in Sound Forge to prepare the test from your files, and the only time I'll ask you to look away is when I rename the files prior to playing them.

Beware the dither in sound forge! It have(had?) faulty dither levels in most of the available options. At least the older version I've tried. Dithering a sinewave and checking the spectrogram after truncation will reveal this. May be worth checking before you do any testing. =)


Andreas N

[Ethan Winer]

Originally Posted by Lupo

Beware the dither in sound forge! It have(had?) faulty dither levels in most of the available options. At least the older version I've tried.

I have version 6.0e. What do you have?

When I null the dithered and truncated files I get what sounds like hiss at the expected level way down at the lowest bits. If you can tell me the exact steps to confirm it's dithering correctly, I'll be glad to test that.

Also, are you certain that you tested it correctly? I just searched the Sony Knowledgebase for "dither," and I also searched their Sound Forge forum, and saw nothing about an acknowledged (or not) bug related to dither.

--Ethan

[Lupo]

Hello!

Have used but don't currently have SF. Think it was version 6. Have downloaded the manual on the website instead, to see what the options are, and what they write about them. Turns out we're both correct.

What I did was to create a sine wave in 24 bits, dither & truncate to 16 bit and watch the results on the spectrogram option. All within the program. Most of the options did not remove the truncation noise, there was still semirandom spikes of energy in addition to the sine and the noise. So I figured there was something wrong with the dither!

At that time, I hardly knew what dither was, only what it was supposed to do. Noise distribution was an incomprehensible word. Looking at the manual now, it is obvious that they've put in several noise distribution options that does not give proper dithering. "It's not a bug, it's a feature!" That caught me out and led me to believe that there was something wrong.

The correct thing to say would be that most of the available dither options are not random enough to totally avoid the truncation artifacts. Though you probably knew that anyway.. :)


Andreas N

[BushmasterM4]

Since my original questions started this whole debate. Do I get to come to the dithering tests at Ethans or where ever its held ? And do I get free Air fare, room and board. Oh yea, food too. I just an old guy trying to understand the whole recording, mixing, mastering processes. Im a hobbiest and would love to be in the company of all brains at work on this thread. Please !!!!!!!!!!!!!!!

[peeder]

Originally Posted by Ethan Winer

When I null the dithered and truncated files I get what sounds like hiss at the expected level way down at the lowest bits. If you can tell me the exact steps to confirm it's dithering correctly, I'll be glad to test that.

Truncation has noise, and in a noise-shaped dither, truncation can actually have MORE noise at certain frequencies than the dither will.

To test your dither, simply download the 24bit level sweep file I posted earlier to this thread, and run it through various dithers and truncation. Take the resulting file and visualize it with something handy...I use izotope RX, spectrafoo might be the best (and I think with your interests may be worth the investment)...I understand you're a PC guy so I don't know all the options on it.

[dannygold]

Originally Posted by Watersound

Ok sorry if this is a dumb question: When I complete a mix in Digital Performer at 24/44 I then bounce it down to 16/44 to burn to CD (when not having it professionally mastered). Would DP not be dithering automatically? Do I need to buy a separate dithering program? If so what do you guys recommend? Thanks!

You asked a practical question, and it deserves an answer. DP doesn't truncate when performing a 24>16 bit bounce. Unless you use a dither plugin like L2/L3 or even DP's quan jr on the master fader. But quan jr sounds really horrible, so don't use it.... ooops, I gues I'm taking sides with THAT comment! Really though, it's not a good dither. Anyway, the ONLY time DP dithers other than via plugins is during 24 to 16 bit conversions... if the dither box is checked in the audio menu. If that menu item is checked, 24 to 16 conversions use dither. If unchecked, truncation occurs. I used work for MOTU, so I know that the dither checkbox is a misunderstood feature. I think I've described it accurately.

Danny Gold

[Ethan Winer]

Originally Posted by BushmasterM4

Since my original questions started this whole debate. Do I get to come to the dithering tests at Ethans or where ever its held?

Yes!

And do I get free Air fare, room and board. Oh yea, food too.

Maybe.

I really would like to see this happen. This stuff comes up all the time, and - obnoxious know-it-all that I am - I always offer to host or attend a blind shoot-out. But when push comes to shove nobody agrees to meet! The dither thread over at Lynn Fuston's forum has now faded away, and so have half a dozen others over the past year. C'mon you guys, put your money where your mouth is. Let's have a Gearslutz event where we can put all of this stuff to the test. Dither versus truncate, 24 versus 16 bits, 44.1 KHz versus higher sample rates, my $25 SoundBlaster Live card and Delta 66 versus the most expensive A/D/A anyone here owns, and so forth. It needn't take hours and hours to do either. The more you believe this "hi-res" stuff matters, the more I want you to attend.

--Ethan

[johnlink]

I'm interested in attending. Ethan, what train would I take from Manhattan to get me closest to your place?

John Link

[Ethan Winer]

Originally Posted by johnlink

I'm interested in attending. Ethan, what train would I take from Manhattan to get me closest to your place?

That'd be great! The best train from Manhattan is Metro North to Brewster, NY. It's just over an hour from Grand Central, and trains run both ways once an hour every hour. The Brewster station is about 25 minutes from my house, and of course I'll be glad to pick you up and drop you off. Anyone else from NYC too. And besides drinks on me I offered earlier, a nice dinner is on me too. For everyone.

--Ethan