[Matroska-devel] Re: [Ffmpeg-devel] Adding matroska support to FFMPEG via libmatroska/libebml, in C++ ?

Alexander Noé alexander.noe at s2001.tu-chemnitz.de
Tue Oct 14 16:00:01 CEST 2003

Steve Lhomme wrote:

> 2) This is only for specific audio codecs that have internal framing. 
> I'm 90% sure any hardware/software player that can decode these will 
> have no problem being fed with multiple frames at once. I know this is 
> the case for MP3 at least. And as, so far, we are the ones who define 
> how each codec should interpret data out of a Block. We can change it 
> easily and that will be the standard way in matroska. And that won't 
> prevent anyone from using 1frame=1block. So I see no real 
> compatibility problem here. 

Some hardware decoders HAVE to be fed with 2 AC3 frames at once. Mosu 
said that
the AC3 filters should be made take care of b0rked hardware.

 > 1) Alex's test shows that we do have the smaller overhead. (that was 
not obivous
 > to him before).

It was. But it is a hack, and I did wait for your permission to hack 
matroska. If
you had said before that I can hack it, I would have done it...

 > So now he cannot claim anymore than AVI can > beat MKV. (of course
 > there are probably cases where it would, but not in 90% of cases)

The AVI overhead is constant at 16 bytes per chunk. Matroska has 13 for
a p-frame and 16 for a b-frame. So with video-only, it is impossible.

If you allow the same frame grouping for audio frames in Matroska as for 
would not see any possibility to get the AVI overhead below the Matroska 


More information about the Matroska-devel mailing list