22 Feb
2007
22 Feb
'07
5:06 p.m.
On Thu, 2007-02-22 at 15:53 -0500, Rick Wright wrote:
This is true, but as Folderol wrote, 32 bits should be
*plenty* of
dynamic range for audio. In fact, it has been argued that ~22bits is
sufficient as beyond this you get into the h/w noise floor, hearing
limits, etc.
32 bits of resolution gets you close to the *thermal* noise floor. i.e.
your recording contains noise contained by brownian motion. not very
useful.
32 bits of IEEE floating point gives you 24 bits of resolution and a
bunch of bits for an exponent to allow summing without clipping.
--p