On Feb 4, 2015, at 5:06 AM, Didier Dambrin <di...@skynet.be> wrote:
Then, it’s no-win situation, because I could EASILY manufacture a
bit of music that had significant truncation distortion at 16-bit.
Please do, I would really like to hear it.
I have never heard truncation noise at 16bit, other than by playing
with levels in a such a way that the peaking parts of the rest of the
sound would destroy your ears or be very unpleasant at best. (you say
12dB, it's already a lot)
I totally understood the point of your video, that dithering to 16bit
isn't always needed - but that's what I disagree with.
-----Message d'origine----- From: Nigel Redmon
Sent: Wednesday, February 04, 2015 10:59 AM
To: A discussion list for music-related DSP
Subject: Re: [music-dsp] Dither video and articles
Hi Didier—You seem to find contradictions in my choices because you
are making the wrong assumptions about what I’m showing and saying.
First, I’m not steadfast that 16-bit dither is always needed—and in
fact the point of the video was that I was showing you (the viewers)
how you can judge it objectively for yourself (and decide whether you
want to dither). This is a much better way that the usual that I hear
from people, who often listen to the dithered and non-dithered
results, and talk about the "soundstage collapsing" without dither,
“brittle” versus “transparent" , etc.
But if I’m to give you a rule of thumb, a practical bit of advice
that you can apply without concern that you might be doing something
wrong in a given circumstance, that advice is “always dither 16-bit
reductions”. First, I suspect that it’s below the existing noise
floor of most music (even so, things like slow fades of the master
fader might override that, for that point in time). Still, it’s not
hard to manufacture something musical that subject to bad truncation
distortion—a naked, low frequency, low-haromic-content sound (a
synthetic bass or floor tom perhaps). Anyway, at worst case, you’ve
added white noise that you are unlikely to hear—and if you do, so
what? If broadband noise below -90 dB were a deal-breaker in recorded
music, there wouldn’t be any recorded music. Yeah, truncation
distortion at 16-bits is an edge case, but the cost to remove it is
almost nothing.
You say that we can’t perceive quantization above 14-bit, but of
course we can. If you can perceive it at 14-bit in a given
circumstance, and it’s an extended low-level passage, you can easily
raise the volume control another 12 dB and be in the same situation
at 16-bit. Granted, it’s most likely that the recording engineer
hears it and not the end-listener, but who is this video aimed at if
not the recording engineer? He’s the one making the choice of whether
to dither.
Specifically:
..then why not use a piece of audio that does prove the point,
instead? I know why, it's because you can’t...
First, I would have to use my own music (because I don’t own 32-bit
float versions of other peoples’ music, even if I thought it was fair
use to of copyrighted material). Then, it’s no-win situation, because
I could EASILY manufacture a bit of music that had significant
truncation distortion at 16-bit. I only need to fire up one of my
soft synths, and ring out some dull bell tones and bass sounds. Then
people would accuse me of fitting the data to the theory, and this
isn’t typical music made in a typical high-end study by a
professional engineer. And my video would be 20 minutes long because
I’m not looking at a 40-second bit of music any more. Instead, I
clearly explained my choice, and it proved to be a pretty good one,
and probably fairly typical at 16-bit, wouldn’t you agree? As I
mentioned at the end of the video, the plan is to further examine
some high-resolution music that a Grammy award-winning engineer and
producer friend of mine has said he will provide.
...and dithering to 16bit will never make any audible difference.
If you mean “never make any audible difference” in the sense that it
won’t matter one bit to sales or musical enjoyment, I agree. I
imagine photographers make fixes and color tweaks that will never be
noticed in the magazine or webpage that the photo will end up in
either. But I guarantee you, there are lots of audio engineers that
will not let that practically (using the word in the original
“practical" sense–don’t read as “almost") un-hearable zipper in the
fade go. If they know it’s there, and in some cases they CAN actually
hear it, with the volume cranked, you can tell them all day and all
night that they are wasting there time dithering, because listeners
will never hear it, but they will want to get rid of it. And the cost
of that rash action to get rid of it? Basically nothing. Hence my
advice: Dither and don’t worry about it—or listen to the residual up
close and see if there’s nothing to worry about, if you prefer.
On Feb 3, 2015, at 10:24 PM, Didier Dambrin <di...@skynet.be> wrote:
Sorry, but if I sum up this video, it goes like this:
you need dithering to 16bit and I'm going to prove it, then the
video actually proves that you don't need it starting at 14bit, but
adds "it's only because of the nature of the sound I used for demo".
..then why not use a piece of audio that does prove the point,
instead?
I know why, it's because you can't, and dithering to 16bit will
never make any audible difference.
It's ok to tell the world to dither to 16bit, because it's nothing
harmful either (it only mislays people from the actual problems that
matter in mixing). But if there is such a piece of audio that makes
dithering to 16bit any audible, without an abnormally massive boost
to hear it, I'd like to hear it.
Andrew says he agrees, but then adds that it's important when you
post-edit the sound. Yes it is, totally, but if you're gonna
post-edit the sound, you will rather keep it 32 or 24bit anyway -
the argument about dithering to 16bit is for the final mix.
To me, until proven otherwise, for normal-to-(not abnormally)-high
dynamic ranges, we can't perceive quantization above 14bit for
audio, and 10bits for images on a screen (debatable here because
monitors aren't linear but that's another story). Yet people seem to
care less about images, and there's gradient banding all over the
place.
-----Message d'origine----- From: Andrew Simper
Sent: Wednesday, February 04, 2015 6:06 AM
To: A discussion list for music-related DSP
Subject: Re: [music-dsp] Dither video and articles
Hi Nigel,
Isn't the rule of thumb in IT estimates something like: "Double the
time you estimated, then move it up to the next time unit"? So 2 weeks
actually means 4 months, but since we're in Music IT I think we should
be allowed 5 times instead of 2, so from my point of view you've
actually delivered on time ;)
Thanks very much for doing the video! I agree with your recommended
workflows of 16 bit = always dither, and 24 bit = don't dither. I
would probably go further and say just use triangular dither, since at
some time in the future you may want to pitch the sound down (ie for a
sample library of drums with a tom you want to tune downwards, or
remixing a song) then any noise shaped dither will cause an issue
since the noise will become audible.
All the best,
Andrew
-- cytomic -- sound music software --
--
dupswapdrop -- the music-dsp mailing list and website:
subscription info, FAQ, source code archive, list archive, book
reviews, dsp links
http://music.columbia.edu/cmc/music-dsp
http://music.columbia.edu/mailman/listinfo/music-dsp