13 Jan
2009
13 Jan
'09
10:46 a.m.
Jorn, i'm not that far into it yet, and have no coding experience or
mathematical degree to back me up. :) Just a former orchestral musician and
composer trying to make his way, who has to use his ears, and a slide rule,
to work things out.
As soon as i know what 2nd level ambisonics are, i'll have a better idea
though. :)
Anyway, back to the books........
Alex.
On Tue, Jan 13, 2009 at 6:28 PM, Jörn Nettingsmeier <
nettings(a)folkwang-hochschule.de> wrote:
alex stone wrote:
and
Those
of
the
at
but i
use
etc,
violins,
convolve
is
cut
metre
listener.
over
Jorn, thanks for feedback. I've just tried
one of Fon's amb plugs, and
it repeatedly crashed ardour, so i think i'd better fix that before
going further.
interesting. i've never had problems with those plugins in ardour. which
specifically? and which ardour version?
As for mike bleed, it's not a full multimix
of pseudo mic blend, but
more a 'hint' of signal from adjacent instruments. It's artifical, imho,
to completely remove any resonant blend of adjacent instruments, and i
already have a modicum of success in terms of 'more lifelike' response
using this method. I'm also using orchestral samples here, not a live
orchestra, so i'm keen to explore just how far we can get down the
'real' road, before limitations prevail.
i see.
As an aside to this, the VSL orchestral sample
library team have already
started a project not dissimilar to this, called MIR, so the concept is
not just mine, or even theirs... :)
I knew i was kinda hopeful when i asked about cutting an impulse into
chunks, so i'm not surprised at all.
Now to get this Amb problem sorted out.
yeah, i'd be interested to hear how it turns out. if you find the time,
post your findings to LAU.
fwiw, i'm just working on a somewhat related project. i have a
multi(close)miked recording of an organ concert with three spatially
discrete organs and a few hamasaki signals, and i'm trying to shoehorn
those into a spatially correct and pleasant 2nd order ambisonic mix in
full 3d. i've taken a leaf from your book and i'm applying individual
delays to each microphone to correct the distance to the (virtual)
listening position i'm mixing for, and i've measured the source
positions in azimuth and elevation to be able to pan them correctly.
results are quite enjoyable so far, but i hope to be able to bribe the
organist to play some excerpts for me again, so that i can take a
soundfield recording for reference... the results will be presented in a
paper on LAC 2009.
have you considered publishing your results as well? the lac paper
deadline is still open iirc.
regards,
jörn
Alex.
On Sun, Jan 4, 2009 at 1:32 PM, Jörn Nettingsmeier
<nettings(a)folkwang-hochschule.de
<mailto:nettings@folkwang-hochschule.de>> wrote:
alex stone wrote:
> Ok, this might be a bit of curly question, and as i don't know if
this
is
possible, either valid or not.
The subject is placement, and pertains to orchestral recording.
(My own
> work composed within the box with linuxsampler, from midi in RG, > recorded in Ardour.)
>
> I'd like to place my instruments as close as possible to an
orchestral
setup, in
terms of recorded sound. That is, once i've recorded,
i'd like
to use convolution and other tools to
'correctly' place instruments
within the overall soundscape.
example:
With the listener sitting 10 metres back from the stage, and
facing the
> conductor (central) my 1st violins are on the listener's left. > first violins occupy a portion of the
overall soundscape from a
point
approximately 2 metres to the left of the conductor, to an outside
left
position, approximately 10 metres from the
conductor, and with 8
desks
> (2 players per desk) about 4 metres deep at the section's deepest
> point, in the shape of a wedge, more or less. That's the pan width the
section.
Now as i understand it, a metre represents approximately 3ms, so
calculating the leading edge of the section across the stage as
'zero',
> the first violin players the furthest in depth from the front of stage,
should, in theory, (and i know this is approximate only, as
i sat
as a player in orchestras for some years, and
understand the
instinctive
> timing compensation that goes on) play about 12ms later than those > the front. Using the ears, and
experimenting, this actually
translates
as about
6ms, before the sound becomes unrealistic, using layered
violin
> samples, both small section and solo. (highly subjective i know, only have
my own experience as a player and composer to fall back
on here.)
make sure that you are using different samples for each desk if you individual delays, otherwise you will
introduce comb filtering
artefacts.
but i doubt these delays will have any perceptible benefit.
> A violin has it's own unique characteristics in distribution of
sound
emanating
from the instrument. The player sits facing the
conductor, and
the bulk of the overall sound goes up, at an
angle, at more or less
30degrees towards the ceiling to a 'point' equivalent to almost
directly
over the listener's right shoulder. Naturally
the listener
'hears' the
direct sound most prominently, (both with ears,
and the 'visual
perception' he gains from listening with his eyes.) Secondly, the
violin
also sounds, to a lesser degree, downwards, and
in varying
proportions,
> in a reasonably 'spherical' sound creation model, with the possible
> exception of the sound hitting the player's body, and those in his
> immediate vicinity. (and other objects, like stands, sheet music, > all playing a part too.)
>
> I've experimented with this quite a bit, and the best result seems
to
> come from a somewhat inadequate, but
acceptable, computational
model
> based on using, you guessed it, the
orchestral experience ears.
>
> So i take one 'hall' impulse, and apply it to varying degrees,
mixed
> with as precise a pan model as possible
(and i use multiple desks
to
layer
with,more or less, so there's a reasonably accurate
depiction of a
pan placed section, instead of the usual pan
sample model of either
shifting the section with a stereo pan, or the inadequate right
channel
down, left channel up method.)
phew! ambitious!
> to make this more complicated (not by intent, i assure you), i'm
> attempting to add a degree of pseudo mike bleed, from my 1st > into the cellos sitting deeper on the
stage, and in reduced amounts
to
the
violas and second violins sitting on the other side of the
digital
stage.
All of this is with the intent of getting as as lifelike a sound as
possible from my digital orchestra.
why simulate mike bleed? i thought you were after creating a "true"
orchestra sound, not one including all unwanted multi-miking
artefacts... i'd rather concentrate on instruments and room.
The questions:
In terms of convolution, , can i 'split' a convolution impulse
with
some
sort of software device, as to emulate the
varying degrees of
spherical
sound from instruments as described above?
you could get a b-format response from every place in the orchestra
(with all other musicians sitting there, for damping), and then it with the violin (which would also have to
be shoehorned to
b-format,
simulating the desired radiation pattern).
but if you have the room and the orchestra, you might as well let
them
play your stuff ;)
> So, one impulse (I use Jconv by default, as it does a great job,
far
better
than most gui bloated offerings in the commercial world)
that can
be, by way of sends, and returns, be
'split' or manipulated not
only in
terms of length of impulse, but fed as
'panned' so as to put more
impulse 'up', less impulse 'down' and just a twitch of impulse
'forward'
of the player, with near enough to none on the
sound going back
into the
player.
i'm not sure i understand 100%, but you might want to look into
ambisonics for that. ardour can do it just fine, all you need to do bypass the panners and use fons' AMB
plugins instead. as to target
format, you could use UHJ stereo. if you desire 5.1, you might want
to
consider working in second order ambisonics.
> I've written this rather clumsily, but i hope some of you experts
may
> understand what i'm trying to
achieve here.
> Can the impulse be split down it's middle, separating left from
right,
aurally
speaking, and if this is possible, can i split the impulse
into
'wedges' emulating that sphere i wrote
of, more or less?
no, i don't think so. you will need a spatial impulse response. the
simplest way to obtain one is to use a soundfield microphone (or a
tetramic, for that matter).
> if there's a way to do this, then i'm all ears, as my mike bleed
> experiments suffer from a 'generic' impulse per section affecting
> everything to the same degree, including the instruments bled in. I
> should note here, this is not about gain, but a wedge of impulse, > out of the overall chunk, that
represents a 'window' or pan section
of
the
whole.
i still don't understand why you're after "mike bleed".
I suppose an analogy for the chunk of impulse
idea would be to
stretch a
> ribbon across a stage, and cut a metre out of the middle. That would be
the bit i'd use, as a portion of the whole, in an aural
soundscape, to manipulate, or place, instruments, to a finer
degree, in
> the attempt to create a more realistic '3d' effect for the > That metre along with other cut out
sections of the impulse
soundscape
could
help me introduce a more....'human' element to a layered
instrument section.
yeah, well, *if* we had a way of capturing a sound field completely such a vast area, we would all be very happy
indeed. it can be
recreated
using wave field synthesis or very high order
ambisonics, but
currently
there is no way of capturing it, other than
measuring a set of
individual points in that sound field.
hth,
jörn