[MPlayer-dev-eng] [PATCH] Automatic downmix

[Tobias Diedrich]

Nicolas George wrote:
> Le quintidi 5 vendémiaire, an CCXIX, Clément Bœsch a écrit :
> > Ok I think I get it. So do those values seem fine to you? :
> 
> I don't think we need "magical" coefficient that no one knows where exactly
> they come from. There may be very elegant and attractive acoustical theory
> behind those coefficient, but this theory stops at two practical objections:

It makes sense to have a comment in the code on how the constants
were derived.  But there is nothing 'magical' about them really.

> - Not all surround files were mastered the same way. There are standards for
>   AC3 and a few other formats (and actually, the standard for AC3 defines
>   several sets of downmixing coefficients), but if you get a 6-channels
>   Vorbis file, you do not which standard it follows.

Well, the standard used doesn't really matter that much for
downmixing, the only assumption made is that each channel gets
routed to one speaker through identical amplifier stages with one
global volume control.  Any difference due to differing listener to
speaker distances are ignored.  After all we can't tell where the
listener is sitting anyway. :)

> - MPlayer is a domestic tool, where speakers are placed near the screen
>   where there is place and sound reverberates on the furniture in a
>   completely uncontrolled way.

It's not limited to domestic use.  There is nothing to stop a cinema
from using MPlayer. ;)

> Therefore, I do not think more than one significant digit would be relevant
> for these coefficients.
> 
> If someone calibrated the position of his sofa with regard to his speakers,
> he is always free to add the relevant pan filter himself.

Sure, but I don't see any reason to limit constant precision. While
I'd say that four significant digits are probably more than required
there is also no good reason to limit it to only one.

The normal user is not going to see these anyway.

-- 
Tobias						PGP: http://8ef7ddba.uguu.de