Seems MegaBitMax Ultra leads the way, although it is a commercial product.  Unless Voxengo has developed something better...

Here are test results from an independant survey
http://www.24-96.net/dither/results.htm

and the actual test
http://www.24-96.net/dither/

Just taken fron Elephant help file:

What type of dithering does Elephant use?

Elephant uses gaussian dithering.  Gaussian dithering is known for its smoothness.  As for noise-shaping, Elephant uses algorithm which gives a very high SNR in the lowest part of the spectrum.

Cheers,

andrea

In my opinion this is a hugely flawed test.  These guys are listening for results on normal loudness levels while dithering occurs at levels close to hearing threshold where perception is different.  To put it short, they are comparing apples to oranges.

As for Voxengo, there are no technical problems for me to create a similar 'Ultra' noise-shaping (I know the math behind noise-shaping).  And I will eventually embed it into the future version of Elephant - it will be based both on maximizing SNR and following equal-loudness contour.  Since it seems that nobody really cares about high amount of high-freq oscillation noise-shaping creates, it is possible to make it awfully huge at frequencies above 19kHz thus leading to a better SNR in the lowest frequencies.

Maybe it's just me, but I don't use noise shaping.  It makes hi-hats sound grainy to my ears.

I can't tell the difference either way, but I still use it.

As long as it remains an option I can choose not to use.  I've found that for myself, I prefer broadband noise for the dither.  It is more natural sounding and it works quite well.  The shaped stuff with more high frequencies is fatiguing to me.

That loudness contour shaping might be an interesting twist though