Dither audio signal to 24 bits

<scorey_plonk@waterford.org> wrote in message
news:g3dcGflo6xbv-pn2-O0O33V3tWj44@localhost...
> Also, even if the source
> material is at 16 bit int, shouldn't dither be applied at the end,
> since 32 bit float processing will produce information beyond the
> resolution of 16 bits?

I'm not sure I understand this.  Are you quantizing back to 16 or 24 bits?  If
so, then dither is appropriate.  If not, then no.

> On the Cakewalk Forum, I believe this was answered in the affirmative:
> <http://www.cakewalk.com/forum/tm.asp?m=119201> It's a very long
> discussion, but if you wade through it you get these points:
>
> 1. 32 bit float gives an actual 25 bits of resolution/information.

Yes!  (Assuming IEEE floating point, which almost everyone uses.)

> 2. If no processing has been done, no dither is necessary even though
> the signal has been converted to 32 bit float.

Agreed.  Assuming no processing or scaling, the conversion from 24-bit int ->
32-bit IEEE float -> 24-bit int is bit transparent.

> 3. You should always dither your final output, whether 24 bit or 16
> bit unless no processing has been done.

If you have the resources, then do it.

> 4. If you are going to do further processing, export to 24 bit using a
> non-noise shaped dither at 24 bits.

Yes, unless of course you have the option to leave in 32-bit float.

> 5. The distortion that occurs when you DON'T dither is not necessarily
> audible.

At low bit-depth (e.g. 8-bit) it is quite obvious.  At 16-bit, I'm sure it is
audible in some situations.  At 24-bit, I highly doubt it.  The distortion is
down around -144dB.  The noise of even the best DACs out today is around -125dB.
So this noise would swamp any distortion and mask it pretty effectively I would
think.

> It could be that point 5. is the overriding factor making the
> discussion irrelevant.  But as far as I can tell, to avoid the
> distortion that occurs when reducing bit depths you should dither your
> final output after DSP regardless of whether your source is 16 bit or
> 24 bit, and regardless of whether your destination is 24 bit or 16
> bit.

Because of the inaudibility of the distortion at 24-bit, the practical answer
IMHO, is to dither when converting to 16-bit (regardless of what your input bit
depth was) but perhaps not to bother with 24-bit.

"Jon Harris" <goldentully@hotmail.com> wrote in message
news:2l0fnpF766stU1@uni-berlin.de...
> The short answer is that triangular "sounds better".

A shorter answer is that you need less of it, in fact I'm told 10 to 15 dB
less of it! The whole idea of dither is that it should be the MINIMUM amount
needed to prevent distortion. Dither is not intended to mask anything, it
simply determines the nature of what a bit reduction sounds like.

As for a "dither buildup," it won't sound nearly as bad as the modulation
distortion buildup caused by failing to dither! The reason for dithering to
24 bits is that subsequent processing can reveal the distortion. My
experience has been that things sound crunchy lots sooner when there is no
dithering to 24 bits.

-- 
Bob Olhsson Audio Mastery, Nashville TN
Mastering, Audio for Picture, Mix Evaluation and Quality Control
Over 40 years making people sound better than they ever imagined!
615.385.8051    http://www.hyperback.com

"Bob Olhsson" news:
> As for a "dither buildup," it won't sound nearly as bad as the modulation
> distortion buildup caused by failing to dither! The reason for dithering to
> 24 bits is that subsequent processing can reveal the distortion. My
> experience has been that things sound crunchy lots sooner when there is no
> dithering to 24 bits.

Sure, but that implies that you need to dither your signal when
truncating to 24bits in the first place. Usually, there is enough
noise present that you don't really need to do this. Also, it was my
understanding that the OP asked about a 24bit *output* file, as in
"final master". There should be no subsequent processing involved
then.

--smb

"Stephan M. Bernsee" <spam@dspdimension.com> wrote in message
news:42c34ef.0407082041.1c636dae@posting.google.com...
> ...it was my
> understanding that the OP asked about a 24bit *output* file, as in
> "final master". There should be no subsequent processing involved
> then.

Been to a Best Buy or Circuit City lately? Can you show me new audio players
that don't almost entirely use DSP controls? How about a broadcast studio?

Extensive subsequent processing is a fact of modern audio life!

-- 
Bob Olhsson Audio Mastery, Nashville TN
Mastering, Audio for Picture, Mix Evaluation and Quality Control
Over 40 years making people sound better than they ever imagined!
615.385.8051    http://www.hyperback.com

Yes, you're right of course - I wasn't considering processing during
playback, obviously.

--smb

Well, but if you think about it - if you assume heavy processing after
the "final master", *any* such quality consideration becomes obsolete.
You neither have control over the noise floor nor can you predict in
what way any prior stage will take influence on the final listening
experience, because if there's so much processing involved beyond your
control you cannot even assume the playback equipment is trying to
reproduce your sound 1:1 (whatever that means - I'd call a straight
D/A without any kind of post-processing like dynamic range
compression, EQ, reverb, en/decoding for data compression/reduction a
1:1 process, or at least an approximation of it). So all you can do is
take the 24bit output file as reference and assume 1:1 playback,
right?

I believe that most average playback equipment can't handle more than
perhaps 18-20 bits of dynamic range in a real world situation, so the
whole discussion is moot anyway. You certainly won't be listening to
music that has this dynamic range at home... if you're not living in
the forest or have very tolerant (or very deaf) neighbors... :-)

--smb

"Stephan M. Bernsee" <spam@dspdimension.com> wrote in message
news:42c34ef.0407100344.55129137@posting.google.com...
> Well, but if you think about it - if you assume heavy processing after
> the "final master", *any* such quality consideration becomes obsolete.

 If you'll think about it some more, signal modulated distortion artifacts
don't build up randomly like Gaussian noise. At some point they become
audible as hard, edgy sounding distortion. Our goal is to push that
"breaking point" as far downstream as is possible.
A perfect case in point can be found in video. There is a director's cut of
Lawrence of Arabia available on VHS. The picture quality is so good that it
makes a joke of high definition video claims . Lawrence of Arabia was one of
the last films shot with a 70mm. negative. The difference between this and
films shot with a 35mm. negative is a total no-brainer even on a mass
duplicated VHS tape.

The minute you start processing signals, you are in the same world as making
analog copies. Unfortunately the perfection of digital cloning doesn't rub
off on digital signal processing!


-- 
Bob Olhsson Audio Mastery, Nashville TN
Mastering, Audio for Picture, Mix Evaluation and Quality Control
Over 40 years making people sound better than they ever imagined!
615.385.8051    http://www.hyperback.com

On 2004-07-18 16:24:32 +0200, "Bob Olhsson" <olh@hyperback.com> said:

> The minute you start processing signals, you are in the same world as making
> analog copies.

Uh no, I don't think so. Analog copies are generally *much* worse wrt. 
noisefloor than the requantization distortion that adds up in digital 
signals if you do all the stuff in 32bit floating point. You would need 
a lot of 32fp/24fxp requantization and processing to come even close to 
the figure - you won't probably get there if you digitally process and 
re-record the mix each time you play it.

I really don't want this discussion to escalate (once again) into the 
world of "digital voodoo" and esotheric claims. All I'm saying is that 
for most practical applications, in a DAW environment extra dither is 
not required when you convert a 32bit floating point audio signal path 
to a 24bit integer output file, since the noise present in the channels 
is usually sufficient to do this for you.

As you know, there are all kinds of extravagant claims in the pro-audio 
biz, so if there's a possibility I'd always let the people choose for 
themselves whichever they think (or hear) is their preferred option. 
I'd set the default for the dither option to "off", though.
-- 
Stephan M. Bernsee
http://www.dspdimension.com

Stephan M. Bernsee <spam@dspdimension.com> writes:

> All I'm saying is that
> for most practical applications, in a DAW environment extra dither is
> not required when you convert a 32bit floating point audio signal path
> to a 24bit integer output file, since the noise present in the
> channels is usually sufficient to do this for you.

Doesn't a 32-bit floating-point number have essentially a 24-bit mantissa
anyway? That is, if the dynamic range isn't too high, you should have little
or no requantization. 
-- 
Randy Yates
Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy.yates@sonyericsson.com, 919-472-1124

"Randy Yates" <randy.yates@sonyericsson.com> wrote in message
news:xxpllhbdeup.fsf@usrts005.corpusers.net...
> Stephan M. Bernsee <spam@dspdimension.com> writes:
>
> > All I'm saying is that
> > for most practical applications, in a DAW environment extra dither is
> > not required when you convert a 32bit floating point audio signal path
> > to a 24bit integer output file, since the noise present in the
> > channels is usually sufficient to do this for you.
>
> Doesn't a 32-bit floating-point number have essentially a 24-bit mantissa
> anyway? That is, if the dynamic range isn't too high, you should have little
> or no requantization.

When IEEE floating-point to signed fixed-point, floating point gives you >=
25-bits, so "essentially", yes.