[PD] block size
bsaylor at macalester.edu
bsaylor at macalester.edu
Wed Aug 28 02:43:20 CEST 2002
Sorry - by inlet~ I meant the inlet of the patch/subpatch, not the
block~ object. Say the signal going into the patch is:
1 3 5 7 9 2 4 6 8 0
If there is a [block~ 4 2] inside that patch, the first signal block
that comes from the patch's inlet is
1 3 5 7
the next is
5 7 9 2
then
9 2 4 6
and so on.
The outlet of the patch overlaps and adds these blocks, so if you just
had the inlet connected directly to the outlet, you'd get
1 3 (5+5) (7+7) (9+9) (2+2) ...
from the patch's outlet.
Ben
On Tue, Aug 27, 2002 at 03:51:15PM -0700, J. Scott Hildebrand wrote:
>
> i'm not exactly sure what you mean by the inlet is giving a block
> consisting of the last half of the previous block and the first half of
> the next block. the inlet takes data in and the outlet takes data out.
> there is only an inlet in the block~ object and no outlet. and how would i
> use this overlapping mechanism to do the overlap and add routine for
> real-time convolution? thanks!
>
> scott
>
>
> On Tue, 27 Aug 2002 bsaylor at macalester.edu wrote:
>
> > I'm not sure what the inlet is for, but yes, just put block~ anywhere in
> > a containing window.
> >
> > The overlap factor is most useful for when you want to do FFT operations
> > on overlapping, windowed (e.g. Hanning) blocks. It sort of means that
> > instead of just moving on to the next input block, the inlet~ will give
> > a block consisting of (say the overlap factor is 2) the last half of the
> > previous block and and the first half of the next block. The outlet~
> > then adds the overlapping blocks back together. You need an overlap
> > factor of at least 2 (which seems to be usually good enough) when
> > working with hanning-windowed blocks. For convolution, windowing isn't
> > necessary, so depending on how your algorithm works, an overlap factor
> > of 1 (e.g. no overlap) might work. I think this is all at least
> > conceptually correct; someone correct me if I'm wrong.. :)
More information about the Pd-list
mailing list