Leonid Makarovsky wrote:

Scott Dorsey wrote:
:don't add the volume, but multiply by some number. So it looks more like
:Y(x) = n * x rather than Y(x) = x + n

: Right. If you added, what you would get would be a DC offset.

Then it does make sense to normalize with multiples of 2.

Close, except that it actually makes sense to normalize with multiples
of 1 rather than 2 (i.e. positive integers). And except that other
seems to be saying that in the real world, software doesn't support it.

The best plan solution by far is to normalize at some higher sample
size (24-bit or 32-bit) and then convert down to 16-bit later.

- Logan