I did a test!
Original MP3 is MPEG-2 compressed @ 64kbps and 22050 khz.
First here is "Difference" of just Delossifier in Stereo Tool (zoomed in in Audacity), run on direct Lame.exe decode of original file (16 bit 22050 khz WAV):
Next is "Difference" of just Delossifier in Stereo Tool (zoomed in to same magnification), run on Lame.exe decode of original file + resampling to 44100 khz using Sox:
Finally, here is "Difference" of just Delossifier in Stereo Tool (zoomed in to same magnification), run on Lame.exe decode of original file + resampling to 44100 khz and increase to 32 bit using Sox:
It appears there is not much (if any) difference between bit depths, but there is between sample rates. The question is... is the Delossifier
more accurate at the increased sample rate, or at the original 22050 sample rate? (I.e., is it removing too much from the 22050 audio?)
(I used a free mp3, it's "Sixteen" by Spheric Lounge, available on their Google Drive here:
https://goo.gl/MZvpE6)