Dithering Shootout - UV22, POW-r, MegaBitMax, etc.

Joined
Feb 2, 2003
Hi there,

I'm not usually on this board, but thought I'd stop by to tell you about a little experiment I'm conductiong.

I've set up a blind test (as well as an open shootout) with samples of most of todays main dithering types (as well as the original and undithered files) and added a voting system.

Find it @ http://www.24-96.net/dither/

Everyone's invited to take the blind test and see which dither they prefer. The results so far (also displayed on the site) are pretty interesting.

I hope this is of interest to some.

Regards,

Robin Schmidt
24-96 Mastering
http://www.24-96.net
 

audiowkstation

Well-Known Member
Joined
Jun 29, 2001
Very cool!

I was experimenting with dither quite alot in the early to mid 90's and created my on dither platform insertion. I hope your test has one that is dither free so I can test mine amongst the samples. Will go there now!

Thanks!
 
Joined
Feb 2, 2003
Well no, the idea of the test is to make the dithering working at low levels audible.

What you hear is the low level music dithered to 16bits. You can hear how the noiseshaping sounds, how well the dithering masks the quantization distortion and how far the dynamic range is increased (i.e. how long during the fadeout can you still hear the music).

The boost in gain makes no difference to the dither, all that is done is that the word is shifted by 9 digits.

The dithering in the files wass applied with great care - Philip Goutier of Steinberg, Alexej Lukin of MegaBitMax and I made sure of that. The added gain is no diferent than you turning the volume of your monitor system up 54 dB...
 

audiowkstation

Well-Known Member
Joined
Jun 29, 2001
The boost in gain raises the dither to audible proportions and remember, digital boost is non linear..that is what Doug is expressing.
 
Joined
Feb 2, 2003
>The boost in gain raises the dither to audible >proportions.

Yes, that's the idea.

>and remember, digital boost is non linear

Well, the gain addition of 54dB in this case is a very simple operaton, introduces ABSOLUTELY nothing new and is 100% reversable.

The non-linearity I think you refer to (if I'm right in assuming you're talking about qnon-linearity of quantized vs. analog audio; i.e. a non-linear transfer function.) bears no relation to this addition in gain, as the word is exactly the same, only that the LSB values are shifted by 9 digits. All values have the same EXACT relation to one another that they had before the shift.

To prove my point:

Take an audio file (anything will do) and save it as test1.wav (or any other format, as long as it's uncompressed pcm).

1. add 12 dB of gain using a dedicated GAIN function in your audio editor and save as test2.wav
2. go back to test1.wav and add 6 dB of gain
3. go have a smoke on the porch :eek: )
4. go back to the file test1.wav, add another 6 dB of gain and save it as test3.wav
5. use a fie compare tool or do a nulling test. The two files will be EXACTLY 100% equal.
6. Now take test3.wav again and subtract 12 dB of gain from it. save it again.
7. now compare it to test1.wav and they will exactly match.

(If your tests don't match, check that you used a pure gain function).

OK, this test isn't rocket science, it sounds like it's proving nothing... but it is: As long as you're only shifting bits, your gain boost is 100% accurate (As long as you're not discarding any information, i.e. clipping).

If you take one of the dither sample files from my dither shootout page and attenuate it by 54 dB, you have the raw dithered file EXACTLY as it was created.
 
O

OTRjkl

Guest
24/96 -
Re-run your test using the CraneSong Analog Dither CD and let us know what you think about it....
 
A

Alexey Lukin

Guest
Originally posted by On-Track Recording:
24/96 -
Re-run your test using the CraneSong Analog Dither CD and let us know what you think about it....
I can say. It's not quieter than UV22, i.e. it sounds around 4 dB quieter than a standard white noise dithering. Although it can sound warmer because of its low-frequency rumble.

Best,
Alex
 
Top