honestly upsampling 48>96>768 seems pretty pointless/silly to me
Don't we all engage in pointless activities in our lives?
The point of using such a high samplerate is bypassing the DAC's internal filtering as much as possible (which would use a higher roll-off frequency, less steep slope, decreased stopband rejection at it's highest sample rate), thus allowing my own filters to pass with the least possible intervention.
About the multi-stage upsampling I'm using... first of all, my ears like the results much better than singe stage upsamling. That part isn't scientific at all
But there's also some reasoning behind it.
Especially at 44.1kHz and 48kHz the necessary filter causes a devilish dilemma: better transients? Then thou shall suffer from more pre/post echo!
Better stopband rejection: say goodbye to bandwidth and transients!
For 44.1->48 I'm using the intermediate filter, and have the settings adjusted to have less stopband rejection and slightly better bandwidth and transient response.
The next step (48->96) uses linear filter and slightly less steep filtering. This allows the transients of the first step to get through (as they're mostly outside the current filter band), but at the same time it attenuates some of the stopband imaging leftovers from the first step without causing additional phase issues.
The last step is mainly to bypass the DAC's internal filtering as much as possible.
I could be wrong with my reasoning, but I really like the end result (I'm willing to share my settings if it doesn't bring the disccusion too much off-topic)