Greetings Forumsters -
A very common occurrence in posts here is to either question or make statements regarding how dither choice effects the sound of a track after requantizing it from 24bit down to 16bit. Since all current shoot outs between various dither algorithms that are available on the net are either flawed due to bias by the manufacturer, being obsolete or incomplete, or presenting samples that do not represent real world uses I have now placed the first of what might be a number of shootouts on my website. The idea is that people can a/b/x these files blindly so that we hopefully can come to some more complete knowledge regarding the issue.

Questions I hope to at least start to help to resolve with this test are:
1) is there a consensus (for at least this particular example) as to what dither choice used allows the 16bit file to sound closest to the 24bit original file?
2) is there a consensus (for at least this particular example) as to what dither choice used allows the 16bit file to sound the "best"? Is this any different than what sounds most like the original file?
3) does dither choice effect the overall "tone" of areas which are not disappearing into the noise floor?
4) does dither choice even matter?

You can download the test files via FTP at -
host name: http://www.totalsonicmedia.com
user name: dither
password: dither

If you need an ftp app I suggest downloading Fetch for Mac from
http://
or WS-FTP for PC from
http://www.ipswitch.com/products/file-transfer.asp

;)

One of these files was just truncated from the original 24bit file.
The remaining 11 other files used the following dither algorithms (NOT IN THE ORDER NOTED BELOW!) - with no additional gain changes or processing whatsoever - when requantizing:

Crane Song "Analog" Dither
Samplitude TPDF
Sonoris TPDF w/ Curve 2 (9th order) noise shaping (from SAWStudio native plugin )
Sony HPTPDF w/ Equal Loudness Contour noise shaping (from their DX dither plugin included with CD Architect 5)
POW-R 1
POW-R 2
POW-R 3
Apogee UV22HR "Normal" (autoblack off)
Wavelab Type 2 - noise shaping type3
Waves IDR (from latest instance available in L3 DX plugin ) - Type 1 - Normal noise shaping
Waves IDR Type 2 - Ultra noise shaping

The original 24bit/44.1kHz file that all of these files were made from is available via the same ftp instructions above as "test.wav"
It is approx 18megs - and unlike the excerpts of the test files it is longer at 1:11.

The recording is of acoustic instruments (2 cellos, violin, viola) live in a room and has a long reverberation tail that goes to the noise floor at its end. It is the final 30 seconds of the string quartet Invert performing "Yumeji's Theme" by Japanese composer Shigeru Umebayashi. It was recorded by Paul Evans Mitchell at Excello in Williamsburg, Brooklyn to Pro Tools HD at 24bit/88.2kHz using stock HD192 converters, Calrec & API mic preamps, Neumann Microtech-Gefell. AKG and Audio-technica mics. It was mixed by myself to 24bit/44.1kHz in my home studio entirely in the box using SAWStudio and a few 3rd party processing plugins by JMS-Audioware, Virtos & Elevayta. There was a large amount of room sound used in the mix but additional artificial reverb was provided by Elevayta's "Convoboy" loaded with an impulse response of a cathedral that was upsampled from 44.1kHz to 88.2kHz.

I received a number of other test files using other algorithms than the ones provided in this initial shoorout. Unfortunately I've "disqualified" the files that were provided to me that used the Megabitmax algorithm as they unlike all the others they did not line up correctly with the original file - I'll need someone to rerun these tests if we are to include them. I also have tests run using the Airwindows, Sweetboy RDR, SAWStudio native and Pyramix algorithms - along with numerous other options for the algorithms noted above - but for the sake of keeping this initial shootout more manageable in both download size and in listening time I've limited it to only the 12 examples included. If this test generates enough interest I will provide a second shootout that utilizes these options that got left out of the initial test. Also - if this generates enough interest I might possibly make the complete shootout (and possibly other examples) available as a data CD-ROM containing wav & aiff files.

and -
PLEASE LEAVE YOUR ANALYZERS/DISPLAYS & FANCY SCHMANCY METERS OFF FOR YOUR INITIAL EVALUATION TESTS. LET's JUST USE OUR EARS AND KEEP THE TESTING "BLIND" FOR OUR INITIAL IMPRESSIONS AND THE START OF ANY DISCUSSIONS! We'll have plenty of time to dive deeper - including looking at pretty pictures - later.

So - happy downloading - and let the discussion begin.

Best regards,
Steve Berson