Dithering to 24 bits for mastering?

[vikingbath]

Hi all!

I am about to do the final bounces before getting my tracks mastered. I have been working in Logic Pro 7 and have a question about dithering. Considering that Logic works internally at 32 bits do I need to dither my mixes down to 24 bits to create the best 24 bit files for mastering?

Thanks for any help!

BEN

[Bob Olhsson]

yes!

My experience has been that dithered 24 bit audio can accept more abuse before it gets crunchy.

[Chi-Squared]

If the ME can work with the 32-bit file, there's no need to reduce to 24 with or without dither. Most DAWs can handle 32-bit files just fine, but I'd say just check with your ME.

[jdg]

Dont worry about dithering. Logic doesn't let u export as 32bit float (or import either )..
but its audio engine is processed at 32bit float...
that said.. because 24bit is fixed point. in theory there should be not errors when downsampling from 32bit float to 24bit fixed...

can anyone with more technical knowledge, please confirm or deny this?

[vikingbath]

Thanks for the advice.

Yeah I don't know the technical details of how all these things work under the hood...If anyone know's more specific to Logic that would be tops !

[Cellotron]

My experience is that if we are just talking about a single truncation of 32bit to 24bit then because any distortions from quantization errors are way below the noise floor that the diffference between the two files is imperceptible even when played back on a great quality 24bit DAC. I should note that 32bit data MUST be converted by your DAW app to 24bit prior to having it played back by your DAC in all cases anyway - so no one is actually listening to 32bit audio anyway!

The issues with the introduction of degradations for many DAW apps is if they continously truncate from a higher bit rate to 24bit while passing the data from one signal process to another (the app I use, SAWStudio gets around this problem for the most part by returning and passing a full fixed point 32bit DWORD after computations are done in 64bit fixed point in any of its signal processes) - but again, ime, a single truncation is really not going to hurt anything. Then again, adding dither to the least significant 24th bit when requantizing from 32bit won't hurt at all either (and in fact might help - especially if requantizing occurs numerous times - which often does happen in a DAW when mixing tracks).

Best regards,
Steve Berson

[ed littman]

Originally Posted by Bob Olhsson

yes!

My experience has been that dithered 24 bit audio can accept more abuse before it gets crunchy.

If one does a null test with both a dithered & truncated file against the original file. you'll notice the truncated is more identical to the original than the dithered, with that said, this observation does not mean I disagree with you.
Ed

[Cellotron]

Originally Posted by Bob Olhsson

My experience has been that dithered 24 bit audio can accept more abuse before it gets crunchy.

by "abuse" do you mean further digital processing?? If we're talking about playing it out the DAC to an analog process chain, unless the DAW app is "broken" then the differences should be imperceptible, if we're talking about comparing a 24bit file that has been truncated once (and only once) instead of dithered when requantized from a 32bit file.

Best regards,
Steve Berson

[Bob Olhsson]

I'm talking about further digital or analog processing. If you do nothing further to the audio, you obviously won't hear the difference but in the real world audio gets processed a lot.

I also agree that just sending a float mix file is probably a better idea.

[MattGray]

Originally Posted by Bob Olhsson

I'm talking about further digital or analog processing. If you do nothing further to the audio, you obviously won't hear the difference but in the real world audio gets processed a lot.

I also agree that just sending a float mix file is probably a better idea.

I've got to agree with Bob on this one, I've often found the same thing. If it's a single truncation from 32bit float to 24bit integar it's not often perceivable. When it becomes perceivable is when further processing is done to it. I did a test with a mix engineer friend of mine using 2 sets of mixes, one with 24bit flat dither & the other truncated to 24bit without dither. When I mastered both sets of files (identically) & compared them blindly we both preferred the dithered mix.

Matt

[Spartacus]

Originally Posted by ed littman

If one does a null test with both a dithered & truncated file against the original file. you'll notice the truncated is more identical to the original than the dithered, with that said, this observation does not mean I disagree with you.
Ed

I think the truncated file may sound more identical to the original than the dithered one according to how the "dither" is processing : for exemple if the change of the added signal (dither) is directly related (correlated) to some aspect of the initial signal, it can become distortion! Indeed, don't forget that 32 bit float is 24 bits with 8 more bits of constant variation depth!

http://www.cadenzarecording.com/imag...tingdither.pdf

[beachhunt]

Originally Posted by Spartacus

if the change of the added signal (dither) is directly related (correlated) to some aspect of the initial signal, it can become distortion!

Maybe I'm mistaken, but then by definition it wouldn't be dither... The whole point is to add non-periodic, non-correlated noise, I thought.

[jdg]

ok back to the OPs question about logic.

does any logic expert know if it automaticly dithers its output from the internal 32bit audio engine when rendering to 24bit?

or does it truncate.

[Bob Olhsson]

You can easily set up a demo to learn how dither works. Just pass music through a noise gate set with the fastest attack and release times. Adjust the threshold until the gate is really chattering away.

Then mix in some noise and you'll hear what dither does. If you eq the mids out of the noise, you can hear what noise-shaped dither does.

A lot of mysticism and technobabble gets applied to what is really a very simple principle. Unfortunately it is not unheard of for software developers to implement dither so poorly that truncation does sound better.

[MBishopSFX]

Here's another simple test to hear what different types of dither and noise-shaping sound like on music:

Play any music or sound source from within the DAW utilizing a digital master fader at the final output, but with the fader PRE-dither/noise shaping. The dither/noise shaping plug-in or device MUST be post-master fader for this test to work. Pull the digital fader down to the point where the audio is just barely audible. Compensate with the analog monitor levels so you can clearly hear the music. BE VERY CAREFUL! With dither OFF, play the audio and adjust the digital master fader to the point where the audio chatters and sputters. You are now hearing the effect of Least Significant Bit chatter at the DAC. Now engage various types of dither or noise shaping and listen to what happens to the audio. Even with the application of the simplest dither you'll notice the chatter goes away. You'll easily hear which dither is most effective and least obtrusive with this test. Try different bit depths; i.e.: 16 or 24. BE SURE TO RESET THE MONITOR LEVELS BEFORE PULLING THE MASTER FADER BACK UP WHEN FINISHED!

After listening this way for a bit (pun intended) you'll clearly hear dither or noise shaping right through the music at normal listening levels.

[Robin Reumers]

I must agree on this one. Dither at 24-bit level might be very hard to distinguish from a truncated one, it just gives me more confidence if dithered, and it's true that it's hard to find converters with better than 20-bit performance, however truncating instead of dithering can give artifacts, even above that 20th bit, I'm sure it'll be VERY hard to distinguish both sources with an AB-test, but there's probably SOME degradation in the undithered version, and it just gives me more confidence to dither.

As a response to the question about there being less different to the original with the truncated version, how did you perform the test? If you count the ones and zero's, it would make sense (even if it could sound worse and audibly less close to the original). Dither will toggle different bits (mostly up to 4 or more) above the LSB, but does this mean that it's going to end up way different from the original? If good noise-shaped dither is used, my bet is that it's going to SOUND closer to the original (even if if it wouldn't mathematically). I'm not sure if this whole thing made much sense, but I hope I got the point across.

Robin

[Bob Olhsson]

Dither decreases distortion by a great deal more than the increase in noise level. It doesn't just cover up the distortion.

Also while dither sounds like noise, you can't assume noise will reduce distortion like dither. There is also subtractive dither where the dither is nulled out after the quantization.

[OMU]

Originally Posted by Robin Reumers

If good noise-shaped dither is used, my bet is that it's going to SOUND closer to the original

I'd use a no-noise shaped dither if further processing is to be done considering the noise albeit linear will be very low.

[Robin Reumers]

I agree with you on that, I shoudn't have said noise-shaped dither, regular dither should be fine too. Sorry about that.

Robin

[Arksun]

I would like to add to the original question as I'm in the same position (using Cubase SX3, generating a 32-bit float file but usually end up sending the mastering house 24-bit file)

If dithering to 24bit is better than just truncating, what type of dither is best.

I've heard that its not a good idea to use extreme dithering with lots of noise shaping as it adversly affects what can be done to the .wav with any further processing. So should we be dithering from 32 float to 24 fixed with no noise shaping at all, or is it safe to use noise shaping?

[musicmann]

Originally Posted by jdg

Dont worry about dithering. Logic doesn't let u export as 32bit float (or import either )..
but its audio engine is processed at 32bit float...
that said.. because 24bit is fixed point. in theory there should be not errors when downsampling from 32bit float to 24bit fixed...

can anyone with more technical knowledge, please confirm or deny this?

I'm pretty sure you're right dude. Is it even an option to start a session in 32 bit? Just up to 24 bit I believe. I just encountered this with a client of mine who asked if i could play back 32 bit files.. I checked with MANY engineers on that, in house and at other studios.. All on PT HD systems. 24 fixed point seems to be it..

[musicmann]

Sorry.. I didn't see the Logic reference.. I was talking a bout PT.. oops
I think its the same deal though..

[Lagerfeldt]

You need to know what kind of dithering or indeed noise shaping Logic Pro offers, POW-r#1, 2, and 3.

I wouldn't recommend adding, e.g. POW-r to the bounce, I'd go with the straight (truncated) 24 bit version.

Remember we're *not* talking 16 bit final dithering, we're talking using POW-r on a 24 bit file that's going to be mastered afterwards.

[Bob Olhsson]

Originally Posted by MBishopSFX

After listening this way for a bit (pun intended) you'll clearly hear dither or noise shaping right through the music at normal listening levels.

That was my initiation into hearing the effect of dither. Some of my clients who have hearing damage from playing electric guitar for years can spot a dither change in an instant.

[diogo_c]

Originally Posted by jdg

Dont worry about dithering. Logic doesn't let u export as 32bit float

Moving a little from OP's DAW:

So, Cubase4 lets you export as 32bit float. Should I export it at 32-bit for mastering? If I export as 24-bit, will Cubase automatically do the dithering?

Nice thread btw. thumbsup

[Lagerfeldt]

Originally Posted by diogo_c

Moving a little from OP's DAW:

So, Cubase4 lets you export as 32bit float. Should I export it at 32-bit for mastering?

No, 24 bit.

[diogo_c]

Originally Posted by Lagerfeldt

No, 24 bit.

Many thanks! thumbsup

[jordanstoner]

While I DO NOT know anything about Logic, I can say this about Pro Tools (HD, in this case - LE is... different). PTHD processes (plug-ins, summing, bussing) at 48-bit, and will add dither at every stage when going back to 24-bit depending on which mixer plug-in you have in the plug-ins folder. Without the Dithered Mixer plug-in(s), it truncates.

I bring that up because it is well documented in the manual as well as in the white papers, and I'm sure that if you search for "dither" or "32-bit float engine" or similar in the Logic manual, it will give you the definite answer of whether or not Logic does this automatically. My *guess* is that it's a preference or option somewhere. Same guess for other DAW's.

Point-in-case - read the manual or visit the website - do the homework. Or ask Lagerfeldt, as he's an awesome Logic reference.

[Lagerfeldt]

Originally Posted by jordanstoner

... whether or not Logic does this automatically. My *guess* is that it's a preference or option somewhere. Same guess for other DAW's.

Point-in-case - read the manual or visit the website - do the homework. Or ask Lagerfeldt, as he's an awesome Logic reference.

Thanks mate,

Nope this is not a preference in Logic.

Also, I don't think it's officially documented in any Logic manual or reference online what happens when Logic bounces to 24 bits.

AFAIK any audio processed destructively in the sample editor is truncated - not dithered. What happens at the mix bus when going from 32 bit float til 24 bit in terms of automatic dither is an unknown factor to me at the moment.

I'll report back if I find out any time soon.

[kjg]

Originally Posted by Lagerfeldt

No, 24 bit.

Why is that Lagerfeldt? Why is it preferable over exporting 32 bit float?
File size? Some DAWs not reading 32f?

regards,
kjg