In article , Justin
Ulysses Morse wrote:

Suppose you record live to 2-track at 16 bits and you just make a
single "transaction" where you maybe run an EQ, a gain boost, and a
little peak limiting all in one pass. You're using 24-bit DSP but you
have to stuff the result back into a 16-bit package. Not a real big
deal, your "clerk" rings you up and says that'll be $45.58. You only
have to say "keep the change" once.

But now what if you've got a bunch of different processes to run,
incrementally, that you evaluate before you move on to the next
process? Maybe you're multi-tracking and you're processing each track
differently. There's all kinds of "keep the change" adding up.

You're really making the case for higher intermediate wordlengths. If
you have a 16 bit file, preferably you'll process even higher than 24
bits. Let's take 48 bit for purposes of discussion. Between processes,
however, if you keep "stuffing" it back to 16 as you say, then you
indeed are going to have trouble, especially cumlatively. In the best
case scenario, a DAW would hand 48 bits from one process to the next and
you'd never come back down until delivery. Unfortunately, most DAWs,
even ones that process at 48, hand 24 bit words between processors (some
do allow 32 float to be saved as an intermediate). Also, external
processors, even those that work at greater than 24 bits of precision,
can only receive and transmit 24 bit words as AES and SPDIF etc. only
support up to 24 bit words. So, you dither from 48 to 24 before handing
it off, but don't drop below 24 until delivery, and that clerk won't be
keeping your change.

This is not to say that nothing above 16 bit capture or delivery is ever
beneficial - just that this is more the argument for processing with
longer wordlengths. Simply staring with a 16 bit file to process doesn't
mean you have to go back to 16 after every process you apply.

--
Jay Frigoletto
Mastersuite
Los Angeles
promastering.com