|
|
Jan 25, 2012 at 10:35pm
pfft~ questions (window & mute)
hi
I have a patch which does kliks – ok, no wonder I am doing some 15 pitch-shifters in it….. I am using Gizmo~ inside of pfft~
a few questions in order to optimise
___I was surprised to see that changing the arguments from 4096 4 to 8129 8 increased the CPU load…can someone explain what is going there? I thought that a bigger window and a bigger overlap would make it easier…
___since I have 5 pfft~ (each doing 3 pitch shifts) what i do at start up of the patch is to mute them all and then demute them at different moments (chosen at random) so their cycles are not the same – right? or should be some other rules here?
__for changing the pitch-factor i use very very long line~ objects which i put inside of the poly. is there a difference between such things done INside or OUTside of a poly (in terms of cpu)?
many many thanks
kasper
|
|
|
Jan 28, 2012 at 8:47am
Hi Kasper,
Did you try just increasing your Signal vector size? If it’s something like 64, I’m not surprised you get clicks. With 2048, do you still get clicks?
On a side note, gizmo~ is an interesting beast. It does not use the same amount of CPU depending on the transposition value.
J-F
|
|
|
Jan 28, 2012 at 11:35am
hi
thank you
my I/O and signal vector size is (as always) 512 256
will try bigger values
also starting the different Pfft~ at different moments seems to help. I had no luck with the 4th argument (the phase thing – recommended by PA Tremblay)
yes, bigger transposition values eat more CPU
many thanks
kasper
|
|
|
Jan 28, 2012 at 9:55pm
Hi Kasper it seems logical you need more power with a step value of 8 (or 1/8 of window size) because it means 8 rather than 4 windows are involved in the overlap-add operations. It uses more memory too.
|
|
|
Jan 28, 2012 at 9:56pm
|
