[LAU] orchestra mixing, reverb, and spatialization

Stéphane Magnenat stephane at magnenat.net
Thu Feb 12 05:09:19 EST 2009

> It's not so simple as it looks. The best spaces (for
> orchestral music) do indeed have a simple 'shoebox'
> shape rather than an 'auditorium' one that widens
> towards the back, has a sloped ceiling, etc. But
> they also have quite complex walls, with lots of
> decoration, balconies, curved surfaces, etc.
> Reflections from such structures are not so simple,
> they are almost never specular (as in a straight
> mirror) but include a lot of (frequency dependent)
> scattering as well. Also 2D geometry is a bit too
> simple - ceiling reflections are important as well.

Of course, I'm fully agree that the model we speak about is an 
over-simplification of the reality. However, should we do simplification, 
shouldn't we try to have the most correct over-simplified model? This is a 
real question, I don't know whether the simplification is just too important 
or if by computing the "correct" early reflections on the sides, the ceiling, 
the back of the stage, and the back of the room leads to a better space 

I was considering ceiling as well when speaking about 2D geometry, but if the 
ceiling is not flat you are right, it becomes 3D ; however if we do a single 
reflection this is still easy. I am not suggesting to build a general purpose 
ray-tracer ;-)

> Unless you have a very weird shape (e.g. very long
> and narrow), the first 4 or 5, or up to about 80ms.
> But see above - what is heard (and measured) as a
> single reflection can already be quite a complex
> affair.

Do you have any experience about whether a simplified but realistic model is 
worth its development time?

> None AFAIK. A reflection in a real room is almost
> never allpass - some bandpass would be a better
> approximation. The allpass filters are used to
> add artificial 'complexity' to the all too simple
> models used by most reverberators, they don't really
> correspond to anything physical. They reduce obvious
> comb filter effects that could result from too simple
> reflection models (e.g. just delay), and also improve
> the reverb tail by 'randomising' it.
> Allpass filters could be used to synthesise the
> complexity of any single reflection, but not in
> the way it's done in most reverberators.

Ok, so should we use an allpass filter, we could as well tune it by ear?

> You normally wouldn't use stereo input to such a
> spatialiser, but position each source independently.

So if one has a stereo input of a violin ensemble that spans 3 meters in 
width, one should use two independent sources to compute the reflections? 
What about depth, should one use two sources, one in front and one in back, 
and compute early reflections for both?

Thank you, have a nice day,



More information about the Linux-audio-user mailing list