....

I have never heard the benifits of dither. I simply don't use it at all. I can hear the difference with cable, clock, converters but never dither. I don't dither coming out of the Panasonic 24bit WZ96 converter to 20 bit adats, don't dither out of the Finalizer into 16 bit soundforge, and it sounds great.

It's important to understand that you can not dither after truncation, only before. It prevents distortion rather than just masking it. Properly designed equipment should dithering automatically so I would never just assume there was or was not any dithering.

Another trap is that many workstations, DAT machines, etc only truncate when the audio is actually recorded. This means, for example if you are monitoring the input, you are not hearing the truncation distortion that is going be present when you play the recording back.

isnt it either truncating OR dithering. in other words, you cant truncate WITH dithering.

shouldnt this be left up to mastering? i mean if im mixing down @ 24bits, i should send the best possible rez to the mastering engineer right? i hear jules is having a helluva time over in da UK finding mastering guys who take 24bit masters. whats the story here in the US?

....

mastering analog... the reason i record at 48khz.

i thought you either dither down or you truncate down digitally speaking. basically dither was low level noise to keep the lsb on instead of lopping it off at 16bits creating zero bit low level [below our hearing] distortion. in other words, you cant have your cake and eat it too.

....

so im sure most here know i record digitally but analog EQ just sounds SO MUCH sweeter. have you found any digiEQ that sounds nice? [now digiEQ CAN do surgical things that analog can never do with FIR]

thanks for the dither explanation.

....

Just wanted to say thanks to you guys for the dicussion in this thread. I'm still listening...and learning.

Nice to hear a pro being +ve-ish about digital. Avoiding the religion. Cheers, Brad.

Best,
Marc

The SP on Dither:

Truncation is the removal of unwanted resolution bits. For example going from 24bit to 16bit.

Dither is the process of minimizing the distortion created by quantisation, re-quantisation and/or truncation.

When you convert to digital, quantisation errors are a natural by-product of creating discrete values from smooth analog waves. These quantisation errors take the form of distortion which is harmonically related to the source material. Truncation adds further quantisation errors by simply discarding the LSBs (Least Significant Bits). By adding a carefully controlled amount of random noise (dither) before truncation the quantisation errors are also randomised. So instead of distortion you now have an equal power of white noise in the mix. More recently processes have been developed to move this white noise high in the frequency range where it is less noticeable, this is called noise-shaped dither.

Truncating without noise-shaped dithering is almost always a bad thing. The only time I can think of where you may not want to dither is if you were passing your masters on for futher processing, for example to a mastering house. Sometimes the cumulative effects of dither can build up and produce unpleasant artifacts. For this same reason you would usually only want to dither the main outputs and not every individual channel in your mix.

<< Properly designed equipment should dithering automatically >>

Sorry to be a pedant but I would say that properly designed equipment should give you the option of dithering automatically.

<< shouldnt this be left up to mastering? i mean if im mixing down @ 24bits, i should send the best possible rez to the mastering engineer right? >>

Depends on the equipment you are using. Let's take ProTools as an example. PT will take 24bit in and 24bit out but the mix bus is 56bit. So even if you are making a 24bit master from 24bit source material there will still be truncation occuring from 56bit to 24bit. So in this scenario you may get improved quality by dithering, even if you are only dithering (+ truncating) to 24bit.

Greg

"""PT will take 24bit in and 24bit out but the mix bus is 56bit. So even if you are making a 24bit master from 24bit source material there will still be truncation occuring from 56bit to 24bit. So in this scenario you may get improved quality by dithering, even if you are only dithering (+ truncating) to 24bit.""" -greg

its a shame you just cant output at 56bits. sure you got a large file but at least it would be the "highest resolution possible" to give to mastering...

thanks brad for the 411, trying to learn about mastering as well, not to do it but to be able to better communicate with the mastering person.

....

I understand ProTools TDM is a 48 bit mixer with an 8 bit overflow i.e. a 56 bit chip with 48 bit ports. The TDM bus that connects the DSP chips, the cards are still 24 bit.

There are actually four possibilities for getting out from the 48 bits. Besides truncating and dithering, a third is arithmetically rounding and the fourth is noise-shaping without dither.

Rounding is a tricky subject. I understand that most DSP chips automatically round unless the programmer turns the rounding function off. While rounding sounds better than truncation, it introduces a DC offset into the signal which can rob headroom and get crunchy. Unfortunately I've been told there have been products where dither was implemented WITH rounding by people who didn't know any better.

Some of the older WAVESplug-inssuch as C-1 and Q-10 use noise-shaping without dither to reduce to 24 bits. This apparently saves gobs of DSP power yet sounds quite good. They are harder to use (without really great monitoring) than the newer ones but in my opinion are highly underrated.

<< its a shame you just cant output at 56bits. sure you got a large file but at least it would be the "highest resolution possible" to give to mastering... >>

But who could import a 56bit file? Furthermore, AFAIK, there are no 56bit D to A converters and even if there were and you were using the highest fidelity monitoring system on the planet, most of the bits of resolution would still be way below the noise floor.

As I understand it, the reason so many bits are used is not really for increased resolution but to minimise floating point math errors during processing.

<< the fourth is noise-shaping without dither. >>

I've always understood noise-shaping to be the process of moving the white noise created by dithering into a less audible frequency band. Is there another sort of noise-shaping and if so what is it?

Greg

thats true greg, and from what i hear jules is having a helluva time finding a mastering person who takes 24bit files, maybe you can hook him up with where... he's around this board somewhere.

Hi Alphajerk,

I'd love to help Jules out but I never work with mastering houses. My stuff all goes to dubbing theatres and is butchered by dubbing mixers! :)

Greg

We take 24 bit, at 44.1, 48, 96, no probs
Just send me an email

Originally posted by Greg Malcangi:
<< But who could import a 56bit file? Furthermore, AFAIK, there are no 56bit D to A converters and even if there were and you were using the highest fidelity monitoring system on the planet, most of the bits of resolution would still be way below the noise floor.

Greg

There's audible sound beneath the noisefloor, and it's not unimportant.

Ben

Below 24 bit? C'mon man...

I know it's there, but you can't even hear truncation distortion from 32 to 24 - you're supposing we can hear musical information that far
down?

Originally posted by Brad Blackwood:

I know it's there, but you can't even hear truncation distortion from 32 to 24 - you're supposing we can hear musical information that far
down?

I believe in magic.. (..and voodoo)
If you're referring to musical information as in meaningful information about rhythm, arrangement, words - no.
If you're referring to tonal, spatial and time-domain information, yes.
My primary concern is that I don't think we have technical headroom enough beyond the audible headroom to effectively lock-out all non-musical intereference.

Sure, I don't have a problem with truncating to 24. I don't have a problem with recording a song with a dictaphone, if that's what I have at hand. But I think it's unnecesary to do it unless I have to.

Anyway, people are still working in the red in fear of, well, what? Maybe they need more headroom... :)

(On a tangent, let's just build squashing into the players instead, so that we don't have to. Then morons can squash all they like)

Ben

Originally posted by Brad Blackwood:

I know it's there, but you can't even hear truncation distortion from 32 to 24 - you're supposing we can hear musical information that far down?

True - 24-bits represent signals down to -144dB - kind of like trying to hear a pin dropping in the next town. A little hard to hear, to say the least.

However, it can (theoretically at least) become an issue when you have apps that truncate, or even dither from higher res operations on every signal in a mix and then sum those together (i.e. accumulated noise from dither could become significant - at least that's my theory). From the example given earlier - PT uses 48 bit processing (with 8 bit overflow I believe, since it is a 56 bit accumulator), but many (if not all) processes are actually using two 24-bit words (i.e. "double precision" marketing terms). Since the TDM bus is 24-bits, all plugins must dither, round or truncate before returning the processed signal to the mixer - which could conceivably add up. Whether it really is noticeable in practice may be subjective, but would be more likely in larger, complex mixes with a lot of dynamics. Mathematically there is some validity to being concerned with what happens at lower bit levels (hence mixers and digital processes with higher floating point bit depths have more room to work - esp. with EQs), but audibly, well, if the mix or process sounds great - I say go with it.

My .02,
Dedric

....

and my comment of saying putting it out at 56bit was purely hypothetical. obviously it made little sense as the PT bus is 24bit regardless.

my noise floor is never below 16bit anyways [except IF there is digital silence between tracks]

....

well my verb tails disappear into noise long before the noise floor, i think they are grainy due to not enough dsp or lack of understanding with how reverb REALLY works... hence the much improved psychoacoustics of the 960L over the 480.

Originally posted by alphajerk:
well my verb tails disappear into noise long before the noise floor, i think they are grainy due to not enough dsp or lack of understanding with how reverb REALLY works... hence the much improved psychoacoustics of the 960L over the 480.

What DAW you use (PT?), what bit depths you record at and any external digital/analog pathes to FX/reverbs. Are you talking about noise in the final bounced file, or in the mix?

The fact that you have reverb tails disappearing into noise could indicate either truncation or possibly accumulated dithering noise in the mix - most likely the former, from either a 16-bit external digital path, or from truncation/rounding or bad dither algos going to the final bounced file.

Regards,
Dedric