Dither...OK, I give up...

[Xenu]

....what settings does everyone here use when dithering from higher bit-depths down to 16-bits?

In Audition, I've always used Shaped Triangular, .6 bits, and 44.1kHz noise shaping...not that I have any idea whether this is at all ideal.

Anybody?

[lukpac]

Wouldn't know. I do everything at 16/44, unless it's getting audio from a DVD or something.

[Dob]

I'd like an explanation of the various options myself. I use 1 bit for dither depth, triangular p.d.f., and C2 noise shaping. But, since I'm completely ignorant of the settings, I'm blindly following the recommendations of (God help me) Grant, who (unsurprisingly) claims knowledge in this area.

I guess I'm relying on the suspicion that the various settings are hair-splitting and won't make an audible difference.

[krabapple]

hydrogenaudio.org would probably be a good bet for a definitive answer, if you pick the right forum there. Or else try the prosound web forums.

[Xenu]

Hydrogenaudio is long on explanations and short on answers. I've read a few threads there, and while they're absolutely fascinating...all I really want is for someone to tell me "for what you're doing, xxxx is ideal."

-D

[krabapple]

Hydrogenaudio is long on explanations and short on answers. I've read a few threads there, and while they're absolutely fascinating...all I really want is for someone to tell me "for what you're doing, xxxx is ideal."

-D

Depends on how you ask the question. And of cousre, you're never going to get jsut *one* answer -- unless you ask which is the best music server software, in which case the answer will be Foobar2k, or the best ripper, in which case it will be EAC. ;>

Really, it sounds like you wants not just an 'answer' but also to know *why* one setting is better,
right? And since youv'e gotten bupkiss here, what harm does it do to ask there?

I also recommended the prosoundweb forums -- where the expertise in this particular area might be more concentrated. Nika Aldrich posts to both, and I'm sure he would have a good answer.

Personally I leave Audition at the default dither settings. I realize that for the 32/88 --> 16/44 transfers I do of SACD material this might not be ideal, but so far everything sounds good.

[combover]

....what settings does everyone here use when dithering from higher bit-depths down to 16-bits?

In Audition, I've always used Shaped Triangular, .6 bits, and 44.1kHz noise shaping...not that I have any idea whether this is at all ideal.

Anybody?

I'd wondered the same thing myself. So I created a clip of low-level white noise (rms around -96db I believe) in 32bit and downcoverted using the various dither shapes, grabbing a screencap of the resulting frequency analysis.

Note that the frequency analysis is linear; as such more of the graph is weighted toward the upper octave or so of hearing range - but dither is most useful there as it's less audible.

http://www.flickr.com/photos/75942954@N00/tags/dither/

As far as recommendations go... I would say give C or C2 a try before 44.1, as they aren't full-spectrum and aren't adding as much noise in the frequency band we're generally most sensitive to - between around 1500-3500Hz. I generally dither with triangular pdf at .5 bit.

[Xenu]

Thanks! That sounds...well, very practical.

[krabapple]

Thanks! That sounds...well, very practical.

I think I have an AES paper by Stanley Lipshitz somewhere that demonstrates the superiority of triangular dither...but in case I don;'t find it, it's probably one of these (taken from his website bibliography)

S.P. Lipshitz, J. Vanderkooy and R.A. Wannamaker. 1991. Minimally audible noise shaping. J. Audio Eng. Soc. 39, 836-852.

S.P. Lipshitz, R.A. Wannamaker and J. Vanderkooy. 1992. Quantization and dither: A theoretical survey. J. Audio Eng. Soc. 40, 355-375.

[krabapple]

and here is an interesting thread about dither etc as related to LP transfers

http://www.hydrogenaudio.org/forums/ind ... opic=38016

[digital]

.

I know, I know - this thread is very old, but here is an answer that (I think) you might be looking for anyway:

http://www.24-96.net/dither

cdnav.com

Andrew D.
.

[krabapple]

Have those guys done any security checking, to make sure people weren't voting multiple times?

[digital]

I dunno', but if you try the various options, you'll be blown away at the differences between many of them. Some sounded the same to me, but most others were way different.

Andrew D.

[krabapple]

Well, I can't get the 'original 24-bit recording' to play -- and when I download it and open it with Audition, it shows as a 16-bit/44 kHz recording of silence. So I don't know what the recording is supposed to sound like. The other two reference files have music (also at 16/44). They should explain what they've done with each example -- I presume they've simply dithered from 24 to 16, but some of those samples sound so incredibly bad I have to wonder what went on. I personally have never heard Audition (Cool Edit's descendent) add noise like that, when dithering.

[krabapple]

Found some more info on the comparison -- on Hydrogenaudio, natch. Bold emphasis is mine.

http://www.hydrogenaudio.org/forums/ind ... opic=9192&

KikeG wrote

Here you can download examples of different dither types available at proffesional industry. The samples are amplified so that the dither noise is easily audible. You can also check using a FFT view the different noise shaping profiles used.

SSRC or FB2K dithering is not available, but If I had space enough to host the files, I could upload equivalent test samples to compare, when I had some time.

According to the available pool results, the best dither without doubt seems to be MegaBitMax Ultra. I fully agree with this, even when SSRC strong ATH noiseshaping would be among the best ones. The famous Apogee UV dithering would not.

KikeG is a very reputable and knowledgable HA poster (he developed the WinABX cxomparison tool), so I trust that this has been done well, though I'm going to shoot him a note about the nonworking reference file.

Also, I got the bitrates wrong -- the 24-bit files are indeed 24 bit, according to foobar2k.