This is what Untertow suggested earlier in this thread. Sonar should only dither when necessary. So if you are working in 32-bit float with dither on and export at 32-bit float then no dither should be applied. It should only dither when you export at a reduced bit depth.

No. It SHOULD apply dither after every process, because in a floating point system every calculation process requires a "reduction" back to the 32 bit floating point nomenclature. If you multiply two 32 bit floating point numbers together you end up with a larger-than-32 bit floating point number. Somehow you have to get it back to 32 bit float. If we merely truncate off the last bits you'll have quantization distortion. We need to have some way of keeping the results linear after doing that calculation. That's what dither is for.

Nika

I guess it's time for a review of the basic principles:

1. Dither is required any time a reduction in bits (that means any bits get truncated off the end) occurs. ANY time - but certainly upon reduction to lower bit depths and likely when converting from float to fixed.

2. Dither is an implied characteristic of digital signal processing. It is not safe to think that it is a requirement that you add it, but rather better to think of it that it is implied that it is there, the ELIMINATION of which causes distortion. It is as necessary to a linear digital system as a cable is to connect the guitar to the amp - it is implied that it is there and the ELIMINATION of the cable causes problems. Someone will have to work very, very hard to convince me that there is good cause to NOT have dither. The only applications would be where either A. no bit reduction is occurring, or B. dither is already applied.

3. We do not need to have a COMPLETELY linear system. We only need to have systems that are linear as much as the ear can hear. If distortion is added 200dB below the noisefloor there is no audible problem. Of course different mixes with different processing do make it difficult to calculate exactly when linearity will creep into the audible range, so maintaining COMPLETE linearity is preferred, but it is not required. If I could keep distortion lower than, oh, umm, say, I don't know - 50dB below the given noise floor of the system I predict that distortion will never become audible through practical and realistic mixing and plugin situations in a real world environment.

4. Floating point systems have inherent problems in adding dither, which raises a valid debate of whether to add it at all mid stream. Adding the same dither formula used in fixed point isn't going to do any harm, but may not actually reduce distortion noticeably. There is no question that dither should be added prior to truncation to lower bit depths like 24 bit fixed, but there are difficulties adding effective dither mid-processing. For this reason, some DSP engineers don't bother and say it's simply not worth the effort and the extra DSP, essentially conceding defeat to the distortion problems inherent in floating point systems. Others, however, spend a lot of DSP and energy getting as effective a dither as they can for floating point and use that. (Nuendo I know takes this approach). I have no hard data on who does what. For more on that topic there is a paper aforementioned at www.cadenzarecording.com.

I hope this answers the question?

Nika

Whoah there! That's a loaded question.

When dithering at any point in the signal chain short of the very last bit reduction before the signal gets to our ears the ONLY appropriate type of dither is pure, random, uncorrelated, white-noise dither of an amplitude sufficient to mathematically eliminate quantization distortion. The easiest and least DSP intensive way to do this is TPDF dither of an amplitude of 2 LSBs.

Now, I believe what you are asking is what type of dither to use at that final stage, and for that we have a different measure. The best to use I believe is generally that which reduces noise to lower than the threshold of hearing across the frequencies of the audible range. This can generally be accomplished with noise-shaping more effectively than with dither. I believe POW-r type 3 does this adequately but I don't know of the other options. With 24 bit dither (or noise shaping) this is less important because the noise floor is well below the audible range for any realistic project anyway.

Make sense?

ORIGINAL: SteveJL
Why is this this such an issue all of a sudden?

I'm afraid it comes from anal audio people like me. After we've turned our audio into numbers, its worthwhile to make sure you handle the numbers with the minimum error possible. Its just code, if it can be made more accurate, why not?

All this talk about distortion and all.........and yet Sonar has sounded, and continues to sound, great, right out of the gate.

I agree completely, I've even managed to make a couple of nice sounding recordings with it myself. I also heard some things that weren't perfect in the old days like dither problems in Sonar 2.2 and more recently the 64 bit engine export bug. The fades and verb tails sound better now after the 5.2 patch.

I just don't get why it is suddenly such a big deal.

It certainly isn't the biggest issue facing mankind today. Its just something I enjoy studying and experimenting with. Mad scientist, I guess.

I realize I probably have lower expectations and hearing skills, so, please, no offence is meant.

No, you're probably more interested in making good music. Even early Sonar could make good music sound just fine. When all the math used on the digital audio is exactly right it might even sound a smidge better.

--Bill

Sounds fair enough Bill, and I can see that. And I can understand wanting to strive for better.

It's just that, in all my time around here, I have never , ever, heard widespread complaints/concerns about the sound quality of Sonar, especially Ver. 3 and on, and I have heard a lot of REALLY good recordings processed through Sonar.....some that rival some of the best Commercial releases.

And, I guess, I'm seeing this become such a HUGE discussion, and, again, all the talk about distortion, I just wonder if something has changed (that I'm missing), or if we are possibly chasing smoke in a mirrored room. I do find it interesting though.

ORIGINAL: SteveJL

And, I guess, I'm seeing this become such a HUGE discussion, and, again, all the talk about distortion, I just wonder if something has changed (that I'm missing), or if we are possibly chasing smoke in a mirrored room. I do find it interesting though.

The disccussion was triggered by the bug fixed in 5.2. After that it all became mainly theoretical. And you are right, we shouldn't loose site of the forest because of the trees. It is just one small aspect of the many aspects of digital recording.

Having said that, the issue becomes a non-issue with one setting in the config pages.

UnderTow

It's just interesting to me I guess. It had been a while since my Digital Audio classes with Pohlmann and like you said talking, reading, and discussing a subject help to spur interest (even made me dust off a few DSP books and order Nika's as well. All in all I think I've cleared up a few personal misconceptions and taken alot from it.

In the end were always talking about such small difference that shouldn't hold back anyones creative juices. I'm just dorky enough that I love the theory, but this thread certainly hasn't kept me from getting my Sonar on to the wee hours each night

Thanks to all who participated,
Chris.

ORIGINAL: Nika

This is what Untertow suggested earlier in this thread. Sonar should only dither when necessary. So if you are working in 32-bit float with dither on and export at 32-bit float then no dither should be applied. It should only dither when you export at a reduced bit depth.

No. It SHOULD apply dither after every process, because in a floating point system every calculation process requires a "reduction" back to the 32 bit floating point nomenclature. If you multiply two 32 bit floating point numbers together you end up with a larger-than-32 bit floating point number. Somehow you have to get it back to 32 bit float. If we merely truncate off the last bits you'll have quantization distortion. We need to have some way of keeping the results linear after doing that calculation. That's what dither is for.

Nika

So what I should do? turn off the dither and export as 32bit float? or ignore all that and mix in Sonar better and then master in Audition?
BTW, are all the dither algorithms the same in all hosts? Or every developer implement them differently?
Sorry for making you bored again.

ORIGINAL: SteveJL

I have a question about all this, and I know I'm gonna sound like a dumbass, but I'm really curious (and I bet others are too).

Why is this this such an issue all of a sudden?

All this talk about distortion and all.........and yet Sonar has sounded, and continues to sound, great, right out of the gate.

Sorry, and I've been following this whole thread and try to get my brain around so much of the "higher" level principles, but honestly, I just don't get why it is suddenly such a big deal.

I realize I probably have lower expectations and hearing skills, so, please, no offence is meant.

Thanks.

Sonar 1 and 2 sounded like a$$ especially when pushed.

mark4man said...

[although...with all the posted test results here...has anyone considered bouncing both ways (dither on/dither off) & listening for what the music sounds like? Or is the focus more for concern that the ME will crank up the noise floor after the fact?]

I have exported like a muttafukka madman you know. I have carried out all kinds of tests. And I want to get to the roots with floating point audio systems, since they behave differently from integer systems when using them as digital mixers in a computer.

Anyways....

I have mentioned before in this thread, that I have sent 24 bit album material to mastering houses. One instance was truncated and one was dithered with TPDF at the two LSB's (when tracks were exported from Sonar).

The master turned out sounding less digital in the latter case. I don't care if it ate up a tiny bit of manouvering dynamic room by putting a noisefloor in the 24 bit files inhearable "area", the end product sounds less digital than the truncated post-mastering buddy.

Both these mastering houses were real stuff with real gear.

How's that for "evaluating the effects".


And there is no concern about the dither noisefloor in a 24 bit file as long as the master is mastered to a NORMAL and HEALTHY and MUSICAL level.

But, if you want 2 dB of dynamic headroom in a true Ricky Martin-manner, then you don't have to dither at all. You will trash your audio in much more audible ways and create much worse distortion by just clipping, limiting, compressing - or whatever - the music out of the music.



So, a new bottomline (and this one is really important)!

I think that most of us who are into these issues (which launched this topic), work with/are doing music that is not madly hypercompressed, moderately compressed or even compressed at all. The more people squash the balls out of their music, the less will they hear of the tiny problems, but instead create much more audible troubles.

When questions like this one arise, it is there for a reason. The reason is that we hear things. We hear things we work on. If we decide to do things the hard way and automate fader movements over a track instead of slamming compressors all over the place, well then I guess we automatically feel a greater concern that this effort will translate in the end. Don't we?

Hoping to clarify and make life simple for the less technical here or those who just want to make music that sounds as good as possible with out faffing about selecting and unselecting dither options...

So here are some points

Points (or facts)

1. As far as I know Sonar will NOT dither if there is no need (ie 32bit float export) even if dither is selected.
   
2. Dither "should" be done when reducing any bit depth and Must be done when going to 16bit.
   
3. Dither uses valuable CPU when tracking (and mixing, but latency is important during tracking.
   

Simple Solution:
If you need to conserve CPU when tracking, then turn off dither... otherwise leave it on... ALL THE TIME. but always have it on when exporting regardless of what bit depth you are exporting at.

Reasoning:
Given that Sonar will not dither if it's not necessary, therefore if dither is on and you export at 32bit Sonar will do nothing.. no dither... no need to turn off dither... if you export at 24bit dither "should" be done.. but whether it can be heard is debateable. But Sonar WILL dither to 24bit, this is NOT a bad thing.
If exporting to 16bit Sonar again will dither and your result will sound nice and smooth.
The only reason that I can see for switching off dither alltogether is to save CPU. All the other cases are handled by Sonar itself.

More complicated solution
Given that some here feel that the 24bit pow-r dither is too noise shaped for a "pre mastering dither" (for want of a better word) in a 24 bit export... switching to triangular dither is an option here to make sure that the 24bit export is as noise free as possible for the ME. But most average people will not find this a problem... if it is a problem for you then select triangular dither.

Bottom Line (as I see it)
Preferably export at 32bitFP if the Mastering Engineer can handle it if not, then export at 24bit.
If mastering yourself, in Sonar, then just select the final bit depth you want (probably 16bit 44.1k if using Sonar to master)... Sonar will sort out all your problems for you.
If using an External App.. eg Sound Forge etc use 32bit export.
Basically you should leave dither on all the time unless you are tracking and need to conserve CPU.... it really is that easy...

Hope this makes sense, and those of you who were confused hopefully you can now rest easy..!!!

I could be proven wrong, but it would take someone reliably telling the difference between 24-bit dither on and off in a double-blind test. I know for a fact that my ears can't discern a difference, although I realize I'm still in the process of training my ears.

Well, export two 24 bit files. One dithered and one without. Play it back in another playback host, but not Sonar. If you have more playback hosts, even better. If you can not hear any difference, well, then I guess it wont hurt to dither the file?

Suggestion: Mark a 10 second area of a mix and export to 24 bits. Mark the following 10 seconds and export to 24 bits with dither. Playback in a separate host (wavelab or such), by placing these exports togeteher one directly after the other and pay attention to if you hear any difference at half time of the resulting 20 second file.

If not - dither won't kill your audio.

But of course, sending the ME something in 32-bit float fixes that problem in a much better way, and I have a hard time believing that any mastering house worth its weight would be unable to handle such a file.

That way you put the decision wheteher the audio will leave the DAC with quantization distortion or not in the mastering engineers hands. Any mastering house worth its weight will most likely do some processing in the analogue domain.

ORIGINAL: dcmg

ORIGINAL: UnderTow
Personaly I enjoy all this stuff, all the theory and technicalities of it all but I don't expect everyone to share that interest. .....If anyone has learned anything from this thread, I'm a happy camper.
UnderTow

Not to take away anyone's fun, but I think many would appreciate a clear cut guideline in simple step by step terms.
Cakewalk really should evaluate this entire issue and consider posting a well thought out FAQ outlining common scenarios with the suggested course of action. There should be more than enough info from this thread alone (props to patton, Nika, Undertow and all who bravely took it on )

I agree, cakewalk should have such FAQ's on their products use to help users. But as others indirectly pointed out-cakewalk did say at first in help files that you should'nt leave dither on it is cumulitive and as undertow pointed out with the 5.2 update Ron pretty much said it's okay. So i guess such a FAQ would constantly have to be updated.

Another would be Waves Type 2 Ultra. Very high frequencies used at low levels.
Max Arwood

Thanks everyone, my head is killing me! Special thanks to Dickie for taking some of the CPU load off my brain with his great summery. Time to stop thinking and make some music...

Thanks again,

George