I noticed that, for amplitude envelopes or other amplitude-related things, it is better to
use a-rate opcodes or to upsamp the k-rate signals before sending them to the audio
modules. Otherwise a certain degree of audible distorsion is always present when kr != sr
(more distorsion for fast transients).
Also it is indispensable to use a kr quite lower than sr when using csound in realtime (I
use 1/100 ratio for realtime task) to increase processing speed.

-- 
Gabriel Maldonado

http://www.agora.stm.it/G.Maldonado/home2.htm


rasmus ekman wrote:
> Use lower k-rate.
> If you need higher resolution for better sound quality, *still*
> try to use lower k-rate and upsample to a-rate with interp.
> 
> I know there are people on the list who will jump to any opportunity
> of setting kr=sr, and if that's what someone needs to be happy, he
> could consider fiddling with k-variable optimisation - else the time
> will be better spent just rendering the thing.
> There is very little that you can do at k-rate which makes a big
> difference in performance. The time-consuming work for Csound is
> (i) inside maths-intensive a-rate opcodes (filters, FFT-based), and
> (ii) the k-rate instrument/variable/everything management.
> There is often not much you can do about the a-rate stuff, since
> the signal after all has to be generated somewhere, but you can do
> something about (ii). Eg set kr=50 through 200, which should be