Dithering 101

[Peakly]

Originally Posted by masteringhouse

But yes, dither is about more than sine waves. As Einstein said:

"It would be possible to describe everything scientifically, but it would make no sense; it would be without meaning, as if you described a Beethoven symphony as a variation of wave pressure."

I just tried it again on a mix I'm working on. I truncated one, and dithered the other. The difference is obvious. The truncated version sounds a little ragged, like it's been torn at the edges or something. The dithered one sounds more cohesive, just easier on the ears.

Of course my far and away favorite is the one that's still in 24 bit. I want to put that on the CD.

[masteringhouse]

Originally Posted by Peakly

I just tried it again on a mix I'm working on. I truncated one, and dithered the other. The difference is obvious. The truncated version sounds a little ragged, like it's been torn at the edges or something. The dithered one sounds more cohesive, just easier on the ears.

Post the examples without telling anyone which is which. Maybe we can put this debate to rest. Maybe not ...

Perhaps Jay Or Riccardo could come up with a reasonable test. For example two audio files at 24 bit. The first part of each file would be the original source track at 24 bit, then followed by brief silence, then either the truncated or dithered version 16 bit version brought back to 24 bit. From there a poll could be setup here to see what the distribution of choices would be.

I would also suggest that the source track be clean and at a reasonable average level (K-20 or K-14?) so that quantization artifacts aren't as easily masked.

[Peakly]

Originally Posted by masteringhouse

Post the examples without telling anyone which is which. Maybe we can put this debate to rest. Maybe not ...

OK. I'll post them. I might not be able to get to it for few days, but I'll do it. I don't know if it will put any debate to rest, but the difference is clear to me. I just don't think truncating to 16 bits sounds very musical, and I'm definitely not going to do it for any of my stuff.

[Ben B]

That method leds itself to cheating, should anyone want to. I think Ethan's original method was probably best (switching between the dithered & truncated mixes by using randomly placed edits with reasonable space between them). Users could report their opinions based on time ranges within the file. This would have to be prepared meticulously, using the exact same mix with no level differences at the edit boundaries.

-Ben B

[masteringhouse]

Originally Posted by Ben B

That method leds itself to cheating, should anyone want to. I think Ethan's original method was probably best (switching between the dithered & truncated mixes by using randomly placed edits with reasonable space between them). Users could report their opinions based on time ranges within the file. This would have to be prepared meticulously, using the exact same mix with no level differences at the edit boundaries.

-Ben B

If I raised an EQ by a few db randomly in a song would you be able to know if it was the EQ or something else in the mix? You need a standard measure in order to establish a difference.

[lucey]

Originally Posted by Ben B

That method leds itself to cheating, should anyone want to. I think Ethan's original method was probably best (switching between the dithered & truncated mixes by using randomly placed edits with reasonable space between them).

Huh? Wha? Why? Without a point of reference you'll never detect the difference. And if you can't trust yourself to not cheat then obviously no one else can be trusted either! So it matters not what they say.

All you need to do is take a 24 bit file, make a coupe of dithered 16s from it, line them up over each other, close your eyes and swap between them quickly 10 or 11 times using mute and solo ... bap, bap, bap ... bap, bap ... bap, bap, bap ... then start to A/B and make your choice. Open eyes, and see. Do that a few times and see how you do.

As far as ... "it's not make or break as so many people claim" . Ethan this is just missing the point. Small differences are often all we get at the high end, and yet, small changes do add up to something tangible and valuable. If yours was my attitude in A/B testing anything I'd not have a career at all. My IC eq has 60 chips in it, and changing them all from one part to another does not make or break anything. The stock unit is fine to many professionals. My tube eq has NOS tubes in it that I selected by ear, and they alone do not make or break anything. The stock unit with new russian tubes is well worth the high price tag. My converter is one of many top of the line models, and it makes no big difference. The XLR cables I use make a teeny tiny little difference vs even a patch bay with beldens basic cable, it's not a big deal. A db here or there in each track makes very little difference, but when you listen to the whole record something different is conveyed with one set of tracks over another. What is small? What is huge? Frame of reference.

"Small" is a word that needs a context, as is "huge". "Small" is a word that talks about the speaker more than the topic. So sure it's small, and it's also huge, maybe, but not for you, okay ... so what? It matters not to anyone else what we find to be "small" or "huge", it's relative to our process and discernment.

Tom, as far as level ... the more heavily compressed and limited a thing, especially with clipping, the more obvious some of these differences can be, or not, and they may show up better with nicely recorded dynamic material. Ultimately, good recording and mixing quality is more important than the level of the masters IMO.

[Peakly]

Originally Posted by lucey

All you need to do is take a 24 bit file, make a coupe of dithered 16s from it, line them up over each other, close your eyes and swap between them quickly 10 or 11 times using mute and solo ... bap, bap, bap ... bap, bap ... bap, bap, bap ... then start to A/B and make your choice. Open eyes, and see. Do that a few times and see how you do.

Perfect. That's what I keep thinking. Anyone can do this with their own material, and hear for themselves. No need for me or anyone else to post files - it's clearer to AB through your own gear and your own monitors. It's easy to hear the difference, and then it's just a matter of deciding what you like. No way does a truncated file sound like a dithered file, and dithers are different, too.

Do any pro MEs truncate at the end of all that meticulous mastering? I can't imagine it, after hearing what it sounds like. But maybe in come circumstances it is done...

[masteringhouse]

Originally Posted by lucey

Tom, as far as level ... the more heavily compressed and limited a thing, especially with clipping, the more obvious some of these differences can be, or not, and they may show up better with nicely recorded dynamic material. Ultimately, good recording and mixing quality is more important than the level of the masters IMO.

Agreed Brian particularly when clipping, though some of the distortion can be more easily masked in a dense mix. I feel that the nasty side effects of quantization distortion are contrasted more in "organic" types of music than say a Metal or Punk song that has been brought up to extreme level.

In some cases some may even argue that additional distortion is s good thing. I once had a Metal band that ran their final mix through a distortion pedal for effect, as well as clients that purposely severely clip and/or truncate individual tracks or parts of songs for creative reasons or to add additional harmonics. In these cases truncation might be seen as "sounding better" in a comparison.

[Ethan Winer]

Originally Posted by lucey

Without a point of reference you'll never detect the difference.

Well, in my test there certainly is a point of reference - the tune itself. Had I posted a file and said, "Guess if this entire file was truncated or dithered" you'd have a point. But I turned the dithering on and off right smack in middle of musical phrases, and so far nobody can tell where I did that.

And if you can't trust yourself to not cheat then obviously no one else can be trusted either! So it matters not what they say.

This is a good point, and I know some people will lie (and have lied in this and other threads) just to prove a point. But in the larger picture it's not so much about lying to prove one's self right. Rather, people are simply mistaken as to what they think they're hearing. When someone claims that truncating from 24 bits to 16 bits "doesn't sound very musical" they need to prove that by identifying the "not musical" parts of my example file!

To paraphrase a post made a few days ago in another GS section, "Moving the mic one inch will give more improvement than using dither."

--Ethan

[Francis Vaughan]

Something I had not realised about Ethan's recording until much later. He still seems to insist on using his ancient Soundblaster Live! for both recording and listening. This card is known to only have 15 bits of real (measureable) resolution. Thus Ethan is not capable of recording any audio with enough resolution to make dither meaningful, and any tests anyone posts with dither or not, his listening environment is intrinsically incapable of resolving any difference. He can't hear the dither because it got chopped off by the DAC before it got out of the sound card. So there is no surprise when everyone else can hear the difference on more modern and capable DACs.

I would encourage Ethan to download a copy (free) of the Rightmark analysis tool (RMAA) and check the test results of a loopback on the card. The Soundblaster Live! is a very B grade device. (Worse, it only uses 48kHz internally, and any other sample rate is forced to transit its internal ancient and poor quality sample rate converter. So if 44.1kHz audio is used it is further degraded.)

(Very annoyingly the RMAA web site would appear to have been hacked - the top page is now a malware fake virus detector downloader. - It claims to have detected all sorts of bad XP viruses on my Mac before insisting on downloading a fake virus fixer - one which is known to be a keylogger trojan. ) Groan.

[lucey]

At mastering there are no mics Ethan, and debating "small" vs "huge" as I said above is just silly. Dither matters, as does 24 to 16. It is not make or break, but it matters. So does 1 db here or there, etc. To nullify the importance of so called "small" differences is a bad way to go. To glorify them also bad. Semantics aside, there are differences, and they are perceptible. So what's that to you? Why do you need to be right here?

Originally Posted by Ethan Winer

Well, in my test there certainly is a point of reference - the tune itself. Had I posted a file and said, "Guess if this entire file was truncated or dithered" you'd have a point. But I turned the dithering on and off right smack in middle of musical phrases, and so far nobody can tell where I did that.

I didn't listen to the test, as it's missing the point. It's all about hearing them against each other. Frame of reference means how we move from that reference and back, detect the difference, describe them to ourselves, and name a cause. We need to at least have the power to turn on and off our best listening when we switch. Your test makes that impossible and thus nulls any hope for real results. It's chaos to prove your point. It's a biased test.

This is a very subtle thing, but it's still a real difference. Hearing this is about feeling the low end and/or hearing the depth change from A to B. Skipping around makes it too chaotic to be reliable. If you don't see that I can't say more.

What I can reliably say Ethan, is that if you were here and I did the test my way, with my converters and dithers that I use everyday, I'd happily bet you any amount you'd like that I could not only detect but name the difference between Linear and powR1 dithers, 85 to 95% of the time. And 16 to 24 bits? 95 to 100% Call it "small" or "huge", it's real, and can be heard.

[Lagerfeldt]

Originally Posted by lucey

What I can reliably say Ethan, is that if you were here and I did the test my way, with my converters and dithers that I use everyday, I'd happily bet you any amount you'd like that I could not only detect but name the difference between Linear and powR1 dithers, 85 to 95% of the time. And 16 to 24 bits? 95 to 100% Call it "small" or "huge", it's real, and can be heard.

Of course, I already did this myself.

[TheMarqueeYears]

Small differences that are audible can be important to practitioners of the art.

Is this not at the heart of what Gearslutz is about. The devil is in the detail. It is striving for the final 1% that derives another 100% of satisfaction.

TMY
Rob

[masteringhouse]

Originally Posted by TheMarqueeYears

Is this not at the heart of what Gearslutz is about. The devil is in the detail. It is striving for the final 1% that derives another 100% of satisfaction.

TMY
Rob

Or 80/20 rule, it takes 80% of the effort to get that last 20%. For me this is what separates the pros and those dedicated to the craft of great audio over more mediocre efforts.

[Ethan Winer]

Originally Posted by Francis Vaughan

Thus Ethan is not capable of recording any audio with enough resolution to make dither meaningful

I have a SoundBlaster but it's not my main sound card. I use it only for editing SoundFonts. My main I/O is a Delta 66.

--Ethan

[Ethan Winer]

Originally Posted by lucey

To nullify the importance of so called "small" differences is a bad way to go. To glorify them also bad.

I agree completely with both statements Brian, and perhaps people here overstate what they think is my point. My point is that of all the things clueless newbies need to worry about, dither and jitter and vanishingly small amounts of distortion are very far down the list. That's all. Andreas Nordenstam (lupo) made some good points, and proved to my satisfaction that he can discern the application of dither in a specific test file he posted in another thread. As he told me in an email, "Regarding the small stuff. It's obvious you don't care but I think you should respect that others do." And I do respect that. But not using the right dither algorithm is never the reason people are unhappy with their productions. As Craig Anderton once wrote, "No listener gives a damn what mic pre you used." And I agree with that too.

I'm just trying to keep stuff in perspective, that's all.

--Ethan

[lucey]

Originally Posted by Ethan Winer

I agree completely with both statements Brian, and perhaps people here overstate what they think is my point. My point is that of all the things clueless newbies need to worry about, dither and jitter and vanishingly small amounts of distortion are very far down the list. That's all.


I'm just trying to keep stuff in perspective, that's all.

--Ethan

cool by me

[Francis Vaughan]

Originally Posted by Ethan Winer

I have a SoundBlaster but it's not my main sound card. I use it only for editing SoundFonts. My main I/O is a Delta 66.

Ah, I sort of wondered, but the implication came from other posts. Still it would be interesting to check the resolution of that card too. I'm quite serious that many cards are not living up to their press. The RMAA self test results for the Delta 66 suggest 97 dB of resolution, which does at least get it close to 16 bits. It is OK, and certainly superior to the Soundblaster.

Sorry for the misunderstanding.

[masteringhouse]

Originally Posted by Ethan Winer

But not using the right dither algorithm is never the reason people are unhappy with their productions. As Craig Anderton once wrote, "No listener gives a damn what mic pre you used." And I agree with that too.

Ethan likewise nobody cares about the acoustics of the room that a recording was made in. If it's enjoyable to listen to that's the main concern. i.e. if it rocks it rocks. Great music always surpasses the media, acoustic, and production techniques used. Personally I still love old Jazz and Blues recordings that are scratchy and distorted. Would Robert Johnson be more engaging if recorded at 24 bit 96k in an acoustically perfect environment?

That being the case forums like this and discussions on audio production or acoustics are rendered pretty much useless except for the fact that this knowledge allows one to acquire the skills in order reproduce audio to a given standard. Personally I find this good stuff to know and not dismiss as hype. This info is not in the same class as $1,000 AC cables or magic stones.

[Peakly]

Originally Posted by Ethan Winer

But not using the right dither algorithm is never the reason people are unhappy with their productions.

I'm not sure about that. I recently complimented the ME who mastered a mostly acoustic project of mine because the dither he used worked great for that music. The other mastering pieces fit nicely too, but I noticed the dither, and mentioned it to him in an email.

[Ethan Winer]

Originally Posted by Francis Vaughan

The RMAA self test results for the Delta 66 suggest 97 dB of resolution, which does at least get it close to 16 bits. It is OK, and certainly superior to the Soundblaster.

Are you talking about input or output? For output, even a 16 bit project exported from a DAW as 24 bits will have 24 active bits because of the math as tracks and effects are added and mixed together. I'm pretty sure both my Delta 66 and SB sound cards can achieve 16 real bits for playback. Even for recording the SB "claims" to handle 24 bits. I put "claims" in quotes because I never tested this. But it can record in 24 bit mode. Versus the old SB Live cards from five years ago that maxed out at 16 bits. Again, I'm not an advocate for Creative Labs, just trying to keep things in perspective.

--Ethan

[Ethan Winer]

Originally Posted by masteringhouse

Ethan likewise nobody cares about the acoustics of the room that a recording was made in.

Good come-back and I mostly agree. Though we've all heard recordings that suck due to bad acoustics. Last year someone gave me a DVD of a live Bruce Springsteen concert that sounds so terrible I gave it back to him.

This info is not in the same class as $1,000 AC cables or magic stones.

Yes, I agree with that too.

--Ethan

[Francis Vaughan]

Originally Posted by Ethan Winer

Are you talking about input or output? For output, even a 16 bit project exported from a DAW as 24 bits will have 24 active bits because of the math as tracks and effects are added and mixed together. I'm pretty sure both my Delta 66 and SB sound cards can achieve 16 real bits for playback. Even for recording the SB "claims" to handle 24 bits. I put "claims" in quotes because I never tested this. But it can record in 24 bit mode. Versus the old SB Live cards from five years ago that maxed out at 16 bits.

The RMAA loopback is pretty much what it says - it measures itself by sending to its input a signal from its output and performing some reasonably sophisticated analysis.

"Claims" is exactly the right word. There isn't a converter on the planet that can achieve a real 24 bits, most manage 21 at best. And that is a best case under ideal circumstances.

The SB Live! was sadly a quite flawed card, and is known to be unable to reproduce 16 bits. The newer ones are better. The 24 bit cards are good, but although the DAC/ADC are 24 bits wide they are hampered by the design of the system and the appalling electrical environment they work in, and thus fall well short. Often only achieving a final real performance similar to an ideal 16 bit system.

As always, the devil is in the details. Simple single number figures of merit never convey the reality. One needs to look at the test results, look at the spectra, look at the environment where they are used, and so on. Internal card based converters are always going to be constrained by the physical environment they work in. And so often they are not especially well engineered from the outset. The difficulty is managing noise and interference compared to a proper standalone converter is always going to mean that a loopback test is a an absolute ideal. Perhaps a more realistic test would be a loopback that transits through a piece of your external gear - like a compressor or equaliser set to bypass.

The other problem is getting past the operating system. So long as you use the DAW tool to directly drive the card and avoid Windows' KMixer and other nasties you are safe. But use Mediaplayer, iTunes, or any other simple tool that does not explicitly bypass the operating systems audio control, and you lose a huge amount of control and quality. XP's insistence on applying its own poor quality sample rate converter to any audio not at 48kHz is enough to nullify most attempts at critical listening.

Again, I'm not an advocate for Creative Labs, just trying to keep things in perspective.

Considering this weeks' self inflicted wound from Creative I wouldn't be advocating anything in that direction.

[Ethan Winer]

Originally Posted by Francis Vaughan

"Claims" is exactly the right word. There isn't a converter on the planet that can achieve a real 24 bits, most manage 21 at best.

Yes, of course I understand that. I just meant that it can record and play back 24 bit Wave files.

So long as you use the DAW tool to directly drive the card

Right, Sonar and Sound Forge can do that.

Considering this weeks' self inflicted wound from Creative

Yeah, that's a pretty amazing story! What a bunch of maroons.

--Ethan