Matthew Smith wrote:
So how would that work? Would the samples then just
end up having 0 for
the 8 least significant bits? (Assuming I've recorded at 24 bits.)
If you normalise (as requested) the samples your 16-bit samples will be
'stretched' (in magnitude) to cover the 24-bits spanning 0 to 1 in
floating-point - hence those bits will be non-zero.
HTH
R