[FFmpeg-devel] [PATCH] swresample/options: change rematrix_maxval default to 1.0

[Michael Niedermayer]

On Fri, Nov 06, 2015 at 01:48:59PM +0100, wm4 wrote:
> On Fri, 6 Nov 2015 12:58:23 +0100
> Michael Niedermayer <michael at niedermayer.cc> wrote:
> 
> > On Fri, Nov 06, 2015 at 11:16:49AM +0100, wm4 wrote:
> > > On Fri, 6 Nov 2015 10:54:28 +0100
> > > Michael Niedermayer <michael at niedermayer.cc> wrote:
> > >   
> > > > On Fri, Nov 06, 2015 at 08:52:32AM +0100, Nicolas George wrote:  
> > > > > Le sextidi 16 brumaire, an CCXXIV, Michael Niedermayer a écrit :    
> > > > > > iam with whatever default people prefer    
> > > > > 
> > > > > IIRC, the current default yields different results when rematrixing from
> > > > > float to float and then converting from float to int than when doing both in
> > > > > a single step.    
> > > > 
> > > > yes, OTOH float -> float rematrix + "inverse" float->float should be
> > > > closer to the original than with the changed default
> > > > The difference between the defaults is in volume of the output
> > > > 
> > > >   
> > > > > I beliee this is not good.    
> > > > 
> > > > no its not good, though the alternative isnt really good either.
> > > > All choices have some kind of disadvantage. There also would be the
> > > > possibility of implementing 2 pass mode to maintain the volume as much
> > > > as possible without cliping (but that could not easily be default)
> > > > For the default, the question is what people prefer ...  
> > > 
> > > Do you mean some kind of DRC?  
> > 
> > no but that would be another possibility as well
> 
> Then I don't know what you meant by this 2 pass mode. Can you explain?
> I'm interested in obtaining better downmix.

i just meant a 2pass mode that would linearly scale the output so
volume is maintained between input and output as good as possible
without cliping (that is apply the same scaling factor to all channels
and all points in time)

dynamic range compression would apply some non linear scaling and
thus strictly speaking add distortion / vary the volume compensation
over time, amplitude or frequency.
drc is better if maintaining absolute volume is more important than
relative volume

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

DNS cache poisoning attacks, popular search engine, Google internet authority
dont be evil, please
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20151107/a258a716/attachment.sig>