Don't know if I want to get into this........
You should dither each time there is a bit reduction. Plugins operate at 64 bit. There should be dithered applied to reduce 64bit data to 24bit (if sonar is set to 24bit). Once again when you bounce to file or master to 16 bit CD. Reducing all data bits at once with 1 time to dither might be best; 64 bit all the way to 16bit, or whatever the final bit rate is to be.
Max Arwood

I 2nd that, but taking care to everything right is important. The right mikes, mike placement, room acoustics, compresion, reverbs Eqs...... Not much difference but it all adds up. That's why some recordings sound better than others
Max Arwood

Cgar,
Yes the internal SA's where set at 2048 FFT.

The External FFT was 4096 FFT. For the original External FFT it was playing back a 24bit sinewave, then that same sinewave was converted to 16 bit with no dither applied. You basically see the same result. The noise floor basically didn't move because the level on the input was cranked up pretty high. What I was expecting to see was the amplitude of the harmonic distortions to raise, but they didn't.

I really think the later graphs show more of what is going on, where it was shown each time you apply dither you basically add +6dB of gain of additional broadband noise.
This is shown in this graph where dither was applied 3 times.
http://www.stashbox.org/uploads/1145924529/16-bit-to-8-bit-comparison.jpg
Look at the purple multi dither graph compared to the single dither graph (yellow).

Then this graph shows where I applied dither during the bit reduction and then applied it in 2 different steps where it was applied after the bit reduction had already taken place and got the same result.
http://www.stashbox.org/uploads/1145926381/Seperate-Dither.jpg

My recommendation is to render at 24bit or 32bit coming out of Sonar when you are mixing. Thus, if you're working at 32bit float and render to 32bit fixed, how much bit reduction are you really doing to need dither applied, as long as your audio signal is peaking near 0dBFS? If you render at 24 bit and your DACs are 24bit, then that's really the same thing you are hearing, since your DACs are 24bit fixed, and so is the rendered 24bit file. Then do your mastering and as a last step when you have to go to 16bit for a CD destination then apply the dither where it really matters. To me this seems like the best path since you are working at higher bit resolutions, where you will have the least noticeable side effects of the bit reduction and then you are only applying dither once, so therefore you are not unnecessarily raising the noise floor with accumalative broadband noise due to the dither noise being applied multiple times.

The basic summary of all this is what SteveJL posted way back when:

Dither is NOT required for monitoring, as the output file is 24-bit and it goes through 24-bit DACS, PERIOD.

Dither is ONLY required during the FINAL Mastering stage.

Dither should ONLY be applied ONCE through the whole process, as it is distortion, and is cumulative.

Final PERIOD <g>.

In this discussion, I still don't believe those points have been proven wrong, but I am open to see it when someone comes thru with some solid backing data proving it wrong.

If you're really going to get anal about all this bit reduction stuff, you should think about asking Cakewalk for a feature suggestion that allows you to render in 32 bit (float) like Sound Forge does. That would probably compliment that new 64bit double precision floating point math of Sonar 5 very nicely that I've been hearing so much about huh?

http://www.stashbox.org/uploads/1145952929/SF-Render-option.jpg

1 to anyone who has ever actually heard the difference, in a full, mastered mix, between having dither on or off for every individual truncation throughout the process.

I have released one album where audio left Sonar in 24 bits with no dither. These files were going to a mastering house that was doing the tweaking with outboard gear (very very good stuff from API, Massenburg, and so on) after leaving first class converters. Even though the mixes were good and everything, the over-all sonic feeling of the end result is a bit cold and certainly a bit digital. It sounds a bit cheap evethough there were some 1300 § spent on the mastering (which took some 11 hours for 11 tracks), eventhough the gear in use was top-notch and everything.

Then I released another album. For these 24 bit exports I choosed to add TPDF at the lowest bits of amplitude of the 24 bit word. This was before Sonar 5, and I used a separate plugin on the mixbus, where the whole mix was "collected". The important thing was to keep the "master fader" at this bus at unity gain since there could not be any process after the dithering apart from the very rendering from float to fixed.

Now - these files went to another mastering house so it is hard to really make a scientific comparison with the first, but all I can say is that those mastered tracks sound less digital. So, was this master better carried out? Could have been. Was the basic mixes different - not really, since my mixing environment and monitoring was the same during the mixing of both these releases. Did they end up completely different from the previous album? NO! They did not! But they sound less digitally focused.

The difference is in whether the mastering house had to tweak 24 bit audio that contained quantization distortion or 24 bit audio that contained dithering noise at the LSB's level of a 24 bit integer word. The mastering process with it's eq movements, it's possible compression and limiting, it's possible blend of digital and analogue editing was putting either:

all kinds of different quantization distortion into a subtle daylight

or

a flat (I'd really like to relate it to linear because of it's nature) dither noise at the lowest bits level of a 24 bit word into daylight.


All I can say is that I really prefer the mastering work that was carried out on the dithered files, and this is only from a pure "microsonic" standpoint.

Was the 24 bit files ruined due to dithering them? No. They sing and swing and sound very, very intime and warm. Do they sound noisy due to the dithering at 24 bits? No. They sound clean as clean can be. What makes them sound different - after the mastering process? Well, I guess there is a difference in putting as much musical information as possible into the sourcefiles, and kick down the artifacts when it comes to mastering a 24 bit integer file in a professional environment. Mastering with outboard gear, top-noth converters and all that will reveal all the tiny crap that has been embedded within the source files.


So bottomline here is; I had to send 24 bit files of two different album projects to two different mastering houses (since they did not accept floating point files). One of these houses received files dithered at the 23:d and 24:th bits level of worth. And this mastering sounds less digital, more friendly and analogue. The "pre-dithered" does not suffer from the dither noise prior to the mastering tweaking, instead it has got more intactness in the body, it's sounding fuller and it really to my ears feels like more music could enter the mastering processors at the time everything happened.

It is naturally hard to really compare things scientifically in this scenario but in the end - the 24 bit files that were dithered did not, by any mean, make the final result suffer from the dither noise.


(Both these album projects were mastered to a level that corresponds with the K-14 system, so they both were steered towards and around the same reference point when it comes to final dynamic headroom.)

ORIGINAL: UnderTow
You can export at 32 or 64 bit float in Sonar 5 and obviously, it won't add dither when exporting in this format.

UnderTow

They also open up quite nicely, (32bit IEEE & 64bit IEEE), in Sound Forge 8, for final prep, before being sent to CD Architect or rendered to other delivery mediums.

I would like to point out one very fundamental thing to you since you seem very stuck in your theory about that monitoring chains noisefloors are so high that quantization distortion is not revealed.

When people (developers, professors, whatever) have evaluated dither at the 16 bit level rendering - which most of us agree is a MUST - they have actually heard the digital breakup of undithered files. They have played them thorugh a DAC, thorugh amplifying stages and speakers and cranked their amplifiers UP in order to hear it. And they have managed to hear the digital breakup!

Maybe once you learn how to read Pattor, then I can take you seriously. Never ONCE DID I negate the benefitS of Dither. Go back and read you dumb ass.

YOU ARE TOO MUCH MAN!

Sonar allows rendering to 32 bit float! What is all your previous bogus about 32 bit integers? We're talking about Sonar and what Sonar is able to do here! We are talking about a floating point wordlength of 32 or 64 bits and we are talking about 24 bit integer files. The 32 bit integer files have no space in this discussion since that format is not anything that has to do with Sonar.

I'm too much? You're the one who started this thread. What's the entire purpose of this thread, if you can save aT 32 bit float. Obviously this was not something I was aware of in Sonar. Guess what I don't use Sonar very much, I use Sony Vegas, Acid and Sound Forge, and there's plenty more things I don't know about Sonar. Obviously you where, so I have to ask what kind of dumb ass would do multiple dithers where it adds noise to the audio signal when you don't have to, and you can wait until the final mastering stage? Well I can now really see 2 of them very clearly who think they're the authority on audio knowledge. Maybe you should stop confusing people? If you knew this then you should be ashamed to even be recommending to apply dither more than once since it is not necessary at all now.

Here I'll stop confusing people now that I realize Sonar can save in floating point format.

1. DON'T USE DITHER AT ALL WHEN MIXING DOWN IN SONAR
2. USE FLOATING POINT SAVE FORMATS, TO AVOID ANY TRUNCATION ISSUES
3. USE DITHER ONLY AS THE FINAL STEP IN YOUR PROCESSING CHAIN BEFORE YOUR FINAL DESTINATION.
4. DITHER IS A FORM OF ACCUMULATIVE DISTORTION, AND YOU WOULD BE A DUMB ASS IF YOU USED IT MULTIPLE TIMES WHEN YOU DON'T HAVE TO.

I think I'm not confusing anyone and those statements are pretty well on the money.

ORIGINAL: Rednroll

I'm too much?

Yes. You are an ignorant fool and a rude arsehole to boot.

1. DON'T USE DITHER AT ALL WHEN MIXING DOWN IN SONAR
2. USE FLOATING POINT SAVE FORMATS, TO AVOID ANY TRUNCATION ISSUES

And what if you need to send your mix to a mastering house that only accepts 24 bit files? Sheesh. Get out of your tiny little deluded world and see what is happening in the real world out here.

4. DITHER IS A FORM OF ACCUMULATIVE DISTORTION, AND YOU WOULD BE A DUMB ASS IF YOU USED IT MULTIPLE TIMES WHEN YOU DON'T HAVE TO.

And again you prove your ignorance. It is really sad that you are incapable of understanding stuff that is being explained to you again and again. Someone once told you that you should only apply dither at the last step and you are incapable of adjusting your ideas in light of new information being kindly provided to you. That makes you a total and utter idiot.

UnderTow

Maybe once you learn how to read Pattor, then I can take you seriously. Never ONCE DID I negate the benefitS of Dither. Go back and read you dumb ass.

Thank you very much for putting the ass-word into discussion. Clearly shows your kind of anger management.

I'm too much? You're the one who started this thread. What's the entire purpose of this thread, if you can save aT 32 bit float.

The purpose is that at some point your 32 bit float must be scaled to fit a 24 bit digital carrier in order for you to hear the music. The purpose was that your music will be slightly influenced by you hearing truncation. The purpose was that you'll most likely hear what kind of actions you take while mixing due to the musical information and not the influence of quantization distortion if there is dither involved during the mix process - during playback and with it enabled when exporting to 24 bit integer files.

Well I can now really see 2 of them very clearly who think they're the authority on audio knowledge.

It's not a matter of authority. It is a matter of implementing facts into a workflow. Does it fit? Does it make any good? Does it matter? An so on. Those kind of questions. This thread was really meant to start a valid discussion about doing this very thing in Sonar. It was never about moving graphs across software platforms. It was about "will it matter if we hear truncation artifacts at a 24 bit level or not" when we CHOOSE what to do whilst mixing.

It's all a fundament for a valid discussion. Sadly, there have not been any attempts to get to the subject.

So...

1. DON'T USE DITHER AT ALL WHEN MIXING DOWN IN SONAR

Yes. Cool. Whatever. Won't ruin your day much. Especially not if your really into static, highly overcompressed music.

2. USE FLOATING POINT SAVE FORMATS, TO AVOID ANY TRUNCATION ISSUES

Yes. Cool. Whatever. You'll save what you hear when quantization distortion influences your mix decisions at a 24 bit level and it is a part of your sound. Your saved file won't include quantization distortion. But when this quzntization later is re-created when rendering the 16 bit file from 32 bit float and with the dithering to 16 bits some of your "sound" might alter, considerably, to your ears.

3. USE DITHER ONLY AS THE FINAL STEP IN YOUR PROCESSING CHAIN BEFORE YOUR FINAL DESTINATION.

Yes. Cool. Whatever.

4. DITHER IS A FORM OF ACCUMULATIVE DISTORTION, AND YOU WOULD BE A DUMB ASS IF YOU USED IT MULTIPLE TIMES WHEN YOU DON'T HAVE TO.

Some digital processors dither the output to 24 bits.

In another host - Sawstudio - there are certain plugins that will always put 24 bit TPDF at the output. If you place one of these native eq's as a plugin it will add dither (TPDF) at 24 bit level on the output. If you place one of these native compressors after the eq, it will calculate the output from the eq (including the dither) and send out a new 24 bit word with NEW dither. Just a eq+compressor on one track will add two rounds of dither to one track in a mix. Do these plugins SOUND good? Yes they sound VERY good! Sawstudio is an integer host, not a floating point host, but it might give you a clue about that dithering at 24 bit levels, and even multiple dihering at these low levels does not hurt that bad that you try to say.

So much for your fears of dither at 24 bit levels. Now, please, understand that we are in this thread only talking about one round of 24 bit dither. A dither that is 7-8 bits below the lowest bit of your final 16 bit word, which, in the end, will be dithered 8-10 bits above the 24 bit level.


Cgar, chris whatever your name is...thanks for actually digging into the very subject ...you got the initital message JUST RIGHT! Just one thing though - we do not want to use noiseshaped dithers (pow-r) when/if we dither to 24 bits, we want a "flat-across-the-spectrum-dither", like the triangular. The options of noiseshaping comes in when dithering to a 16 bit level.

ORIGINAL: UnderTow


I think the whole dither or not issue can be summed up like this:

Whatever the reasons are to dither when reducing bit depths to 16 bit, those reasons are the same and also valid when reducing to 24 bit.

If you think that you need to dither when going to 16 bit, it means that you also need to dither when going to 24 bit. If you think that you don't need to dither when going to 16 bit, then you don't need to dither when going to 24 bit.

The trade-off being made between noise and distortion when applying dither at 16 bits is the same trade-off being made at 24 bit. The only difference is that the levels of the noise or distortion is lower.

UnderTow

I can live with that.

So that people can make a more informed decision, I created a couple of wav files demonstrating the effects of dither on bit reduction. I took the following steps:

I took a regular 16 bit file with a normal signal level (Peak: -1 dB RMS: -15 dB).
I converted it to 32 bit to be able to reduce the gain without loosing resolution.
I reduced the gain by 80,2 dB so that the signal is only triggering 3 bits of the top 16 bits of this 32 bit file.
I converted this to 16 bit. Once using no dither and once with TPDF (triangular) dither.
I then increased the gain by 58,2 dB to make things audible.

These are the resulting files:

http://home.casema.nl/ajohnston/Dither/NoDither.wav

http://home.casema.nl/ajohnston/Dither/TPDF-Dither.wav

If you listen to the non-dithered file, you hear that the signal is heavily distorted and drops out when the signal is triggering less than one bit (out of 16) . This quantization distortion applies to all amplitudes of a signal being bit-reduced. Not just the very low ones in this example. The dropouts will only affect signals that are so low that they don't trigger any bits. (The dropouts in a fade-out that Rednroll mentioned). Of course, the distortion artifacts won't be nearly as loud as this as the file has been boosted by arround 60 dB.

Now if you listen to the dithered example, the original signal is there and not distorted. This is despite the fact that in the 32 bit file most of the signal is actually encoded in bits that get dropped during the bit reduction process. We have now effectively increased the resolution and encoded signals below 16 bits into a 16 bit format file!

These examples also demonstrate that we can hear signals that are below the level of the dither noise. We have a musical signal at an average RMS of -37.5dB in a noise level of average -35 dB!

Keep in mind that the noise you hear in the dithered file will be much much lower in a normal convertion (especially at 24 bits). The same goes for the truncation distortion in the non-dithered file. That major difference is increased resolution and reduced distortion.

I hope this makes it clear why I advise everyone to have dither turned on for any bit reduction process.

UnderTow

ORIGINAL: UnderTow

Hi Nika,

Welcome to this forum.

I don't know if there are any remaining issues. I think most of the stuff has been handled. The only thing I personaly am not entirely sure of is how the quantization distortion is related to the signal frequency. From my understanding the distortion is more correlated to the signal at lower frequencies. Is that correct and how exactly does that work? Or is this purely related to the fact that distortion of higher frequencies are aliased back into the audible band and thus less correlated with the actual higher frequency signal?

Cheers,

UnderTow

UnderTow

Quantization distortion is a form of distortion. Distortion is inherently "correlated" or "coupled" with the material, so that as the material changes the frequencies present in the distortion change. Therefore, as you change the material that goes in (including the frequencies of what goes in) what comes out changes. Therefore, the distortion is ALWAYS correlated to the frequency content.

Nika

ORIGINAL: Rednroll

The main reasons to go from 24 bit to 32 bit float are the increased headroom flexibility

Ohhhhh...you get increased headroom? Hmmmm??? I did not know that, and here silly me always thought that 0dB was always 0dB no matter if it was 8bit, 16bit, 24 bit, 32 bit or 64 bit. I just thought you gain extra values inbetween -inf to 0dB to reduce distortions due to quantization error, by having more values along the Y-axis to get a closer approximation of the amplitude of the audio when sampled? But hey, if you say I get extra headroom that's nice to know. Wait until I let the rest of the guys know, that 0dB is really not the limit in digital audio. Thanks for all that correct information.

Yeah, not really.

Indeed you do gain possible "headroom" from going to larger bit formats in essence. This, because the 0dB point isn't always full scale. Much of the time the some of the extra bits are tacked on to the left of full scale for headroom in processing, such that when taking a 24 bit signal and putting it into, say, a 48 bit system, some of the bits go on the left and others go on the right, in relation to which 24 bits come out of the system.

Also (and this is pretty picky) 32 bit float actually gives you 25 bits of linear data, plus the scaling of 8 bits.

Finally, there are other situations in which signals can exceed 0dB, but those situations are not beneficial.

I'm not really following the conversation, but saw the need to clarify this point, not really knowing what the ramifications of this assertion are.

Nika

Nika...

I have a questions for you.

Since Sonar now can work in a 64 bit float mode, instead of the "ordinary" 32 bit float - will this provide a better "room" for the TPDF (if this TPDF also have this manouvering room) to correctly correlate the quantization errors to the dither noise, or will it still create a scenario where some distortion is "handled properly", while some will just be masked under the dither noise?

ORIGINAL: pattor

Nika...

I have a questions for you.

Since Sonar now can work in a 64 bit float mode, instead of the "ordinary" 32 bit float - will this provide a better "room" for the TPDF (if this TPDF also have this manouvering room) to correctly correlate the quantization errors to the dither noise, or will it still create a scenario where some distortion is "handled properly", while some will just be masked under the dither noise?

Hmm. I don't really like the terminology but I think I know what you're asking. In any floating point system you are bound to have distortion. It is impossible for a dither algorithm to completely decorrelate the signal from the error in a floating point system. All it can do is diminish that correlation. Therefore, distortion is bound to occur in such systems. The more bits you add to the right of the last bit you intend to keep, and the more bits you add to the dither, the more you can decrease that correlation. The result is that you decrease the distortion. With enough bits you can probably reduce that distortion to a level that is effectively infinite because it is too far below the threshold of hearing in any plausible mixing situation to become audible. My personal belief is that in a 64 bit situation, implemented properly, this is the case.

Does that answer the question?

Nika

Hopefully you're learning something here, and that's dither ONCE to avoid noise build up from multiple dithers and wait to dither as the last process.

Hmm. I'm not sure where this is going, but I cannot in any way condone this advice.

Nika