22 Jul
2007
22 Jul
'07
3:29 p.m.
On Sun, 2007-07-22 at 10:48 +0100, Dan Mills wrote:
On Sun, 2007-07-22 at 09:45 +0100, Steve Harris
wrote:
1) the ones that do are using fixed point math, not floating point
2) there are some good theoretical arguments for needing more than 32bit
floating point resolution for a mixer
--p
On 21 Jul 2007, at 13:56, Dan Mills wrote:
Granted.
Be very careful how you write float->int
conversions (it is not
trivial), and work in floating point as far as is possible, there is
little reason (other then marketing) to move to doubles (IIR filters
possibly excepted).
Well, I'd qualify that as there's no need to use doubles between
modules, there are many DSP processes that work better in double if
you can afford the extra memory bandwidth. Filters are one. It's generally not doubles in the audio path
though, but its things
that you're using to modify it, coefficients, phase accumulators and
so on.
I was thinking mainly about some commercial DAWs that boast of stupid
word length accumulators for mix buses and the like. I never did
understand that.