20 Jan
2009
20 Jan
'09
12:34 p.m.
alex stone wrote:
I've done some crude experimenting and ask at this
point, how do i
determine the format (i.e. short b-form ir or not) of a particular
IR? Is there a device or some sort of utility that i can use to do
this?
to clarify things, there are no "short" or "long" IRs. what fons
means
is: the early reflections are the most characteristic aspect of a room,
and they affect localisation the most. therefore, if you want to have
ultra-realistic reverb, use an IR that was measured with the speaker
where your intended instrument is and a soundfield microphone where the
listening spot is. of course, in practice this is not done.
so instead you use one reverb IR instead. it can be short (tail
truncated) to save CPU, because the tail is decorrelated (blurred) and
does not provide localisation cues, hence it would be wasteful to render
it in b-format.
the problem is: you will probably not get such a two-stage reverb
calibrated unless you have a lot of experiences with both b-format and
convolvers.
if you can make it to lac2009, let's talk this through over pizza
(that's how i learned my first steps in ambisonics from fons, and it
works suprisingly well).
an IR can only be in b-format if it contains 4 channels (WXYZ) for full
3D or 3 channels (WXY) for planar surround only.
The reason i ask is, i have a set of IR's for a
hall, recorded 3
front stage, left middle right, and 3 back stage, left middle right.
I built a jconv .conf file with these IRs, and the result was near
100% CPU, and a vertiable flood of Xruns. (AMD 64 Dual Core X2 5600+,
64bit Ubstudio) I understand from this, i've used 'full' IRs, and
Jconv is attempting to process them all at once.
probably, and the cpu usage goes up quickly the longer the IRs are.
also, jconv has the nasty tendency to starve ffado hardware threads (not
really jconv's fault, but needs some tuning), so if you are using a
firewire interface, you might be in for some fun. it can be done though,
works nicely for me, but it requires some poking.
So, I gather from what you've written, that there
is either a
mechanism for creating short b-form IRs from a single IR wav, or i'll
have to find a specific IR that is recorded in such a way that i can
use portions, or sections of the IR, without taking such a
performance hit, or there is a mathematical means of extrapolating
sections from a single ir, already present in Jconv, and i'm not
'getting it', at this point.
as i said, i would recommend against attempting such a split reverb
method, because very likely things will go haywire at some point.
better bribe fons :)
for now, i'd recommend to use one single reverb for all sources, and
only vary the dry/wet ratio according to the distance to the virtual
listening spot. should be really ok.
what is your desired output format? plain old stereo, or 3d ambisonics?