[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 12:01:10 CEST 2003

Steve Lhomme wrote:

> BTW, did you try "6 AC3 frames per matroska Block" in your tests?

With frames of 1536 bytes, you need 7 bytes to code the size of each 
block, so that lacing is not an option.
If you suggest to put several blocks of AC3 audio into one matroska 
block, Mosu will kill you (and me
as well), but I will try that. BTW, a constant-frame-size-flag in the 
block header would resolve that issue.

> It's not the best thing to do because of timecode drifting or edition 
> boundaries. But it's still possible and valid. It's all a decoder 
> problem, not a matroska problem. And as the AC3 (or MPEG audio 1/2/3) 
> decoders are usually done as data stream, feeding them with 1 or 6 
> frames should work the same. At least as well as when data come from 
> AVI... So IMO your test is a bit biased. 

I test standard conform. If you want to see results with violations of 
what we have agreed on, then
I will do those tests and add an option to avimux gui to write such 
b0rked files on the risk of the user.
If you put 6 frames of AC3 in one matroska block, the saved overhead, 
compared to AVI, should
be one byte per frame in theory. It would IMHO be a hack, but you know 
that I like hacks, and I
will implement it if you explicitely tell me to implement a hack.

But see the constant-frame-size-flag-thing for lace headers. It would be 
no hack and achieve the same affect.


More information about the Matroska-devel mailing list