hey, Olaf,

many thanks for pointing this out, this here self on mine stands corrected. *yes*

as there always exists at least one exception which might rather compromise the occasionally erroneously percieved precepts, all my above statements are edited accordingly, hopefully appropriately. *yes*

had some kind of quite strong hunch but did not came to the mind me owns, about hardware limits, this fairly places most of the topic into its rather rightfully deserved position...

the greatest difficulties might come up if, when, one begins to audition the results of the different dithering, noise shaping, algorithms with the volume set at eleven plus, to check the waveforms zoomed in to sample resolution with the view maximised at the zero level point, to hunt for nearly non-existent artifacts... *yes*

all the best for all of us, including the best sounding exceptions amongst our algorithm choices...

 

If you're mixing down to a two-track file inside your DAW, and you're working at (whatever sample rate) and 32-bit float bit-depth, mix it down to that sample rate and bit-depth. That way, you now have a two-track stereo master with your compression, mastering EQ and limiting executed at the highest possible level without any conversion introduced into it. You can then take this stereo master and do all your sample rate conversion and dithering in a two-track editor (or in a different project file in your DAW) and save the converted version as a different file. Always make sure you print your master at the highest quality you possibly can!

You'll want to down-convert your sample rate and bit-depth to CD quality (44.1 kHz/16-bit) before making a regular CD master or submitting it to iTunes, since both pretty much require CD quality to play back properly on most regular .mp3 players and this is the only format you can burn a regular audio CD in. I do all my conversions in Wave Editor on the Mac, but just about any decent two-track audio editor on any OS platform can handle it, or should at least be able to load up the necessary plugs to allow you to do it properly. Wave Editor uses iZotope algorithms (like Ozone), so I usually dither using MBIT+ at the highest possible settings. Sample rate conversion is a far less prickly issue, and I just use the built-in processing for that. I also use the rule of "2" (downconverting quantization from a divisor of 2): I record, mix and master at 88.2 kHz/32-bit float.

There's little difference between 24-bit and 32-bit float, which is why you can play back 32-bit audio on hardware designed only for 24-bit operation. It's basically 24-bits with references to peaks existing outside the 24-bit spectrum.

 

Just yesterday, I found this excellent text that will answer all dithering questions:

http://downloads.izotope.com/guides/izotope-dithering-with-ozone.pdf