view DOCS/tech/oggless-xiph-codecs.txt @ 23014:e7d38f99a8b3

Add perspective projection.

Title       Embedding xiph codecs like vorbis in containers other then ogg
Version     2006-07-30 (draft)
Status      this is not a standard or otherwise accepted by xiph or any other
            group, one day when we have a fully working implementation, did
            enough testing and so on we might submit it to IETF? to become an
            RFC ...
            furthermore this document has been submitted to vorbis-dev and
            so far has been ignored, maybe they where too busy maybe xiph
            wants to prevent their open codecs from being used in containers
            other then their own?
Author      Michael Niedermayer (michaelni at gmx dot at)
License     GPL + GFDL + anything neeeded to turn this into a open standard
            like a RFC

Minimum container requirments:
This appendix only explains how to store xiph codecs in containers which 
support at least one global header per stream, can separate individual codec
packets and in principle support the codec, so for example in the case of
vorbis that would be variable bitrate and variable number of samples/packet
Storage in other containers is outside the scope of this appendix


FIXME non vorbis
Global header:
If the container can store 3 headers per stream in an unambiguos and ordered
way then they shall be stored in that way, if OTOH the container is only
capable to store a single global header then the 3 codec headers shall be
concatenated without any additional header, footer or separator between them
to recover the 3 headers from such a global header the following procedure
shall be used:

1) search for the 1st occurance of 01,'v','o','r','b','i','s'
   the found match and the following 23 bytes are the 1st header packet
2) search for the 1st occurance of 03,'v','o','r','b','i','s' after here
  3) read an unsigned integer of 32 bits and skip that many bytes
  4) [user_comment_list_length] = read an unsigned integer of 32 bits
  5) iterate [user_comment_list_length] times {
       6) read an unsigned integer of 32 bits and skip that many bytes
     }
  7) skip 1 byte
8) the match in 2) and what follows until here is the 2nd header packet
9) search for the 1st occurance of 05,'v','o','r','b','i','s' after here
   the matching part and what follows is the 3rd header packet
if the container needs an identifer for the global header, for example a 4cc
for a global header chunk then glbl shall be used


Storing packets:
Each codec packet shall be stored in exactly one "container packet"
and one "container packet" must not contain more then one codec packet
"container packet" here means the smallest separatable unit of data in the
container


Codec Identifer:
xiph-codec  4-cc id     long id
Vorbis      vrbs        vorbis
Theora      ther        theora
Tarkin      trkn        tarkin
Flac        flac        flac
Speex       spex        speex

if the container uses 4-character codes 4-cc identifer from the table above
shall be used
if the container uses arbitrary length strings as identifers then the long
id from the table above shall be used


Examples and Disscussions about specific containers
What follows are some notes about specific containers, these notes are just
informative as they just repeat what is written above or in the 
specification of the specific container


Example and Disscussion of the avi container
avi supports everything needed to store vorbis, this does not mean that all
application will support vorbis in avi as vorbis is rather different from
other audio codecs commonly stored in avi ...
avi supports a single global header like wav does, the 3 vorbis headers
shall be stored in it and only in it as described above
dwSampleSize must be set to zero as vorbis is vbr, many applications do
this incorrectly for other vbr codecs and consequently vbr audio in avi
becomes problematic
avi does not have timestamps but each chunk has a constant duration, while
vorbis packets can have one of 2 durations, if now the avi header is setup
so that each avi chunk has the same duration as the smaller duration of
the 2 possibilities in vorbis then simply inserting empty avi chunks will
allow every avi chunk to have the correct duration, this is of course
not the most beautifull solution but it is the only way to keep things
exact, additionally note, that empty chunks have been used since ages
in avi to lengthen the duration of video chunks


Example and Disscussion of the asf container
asf supports a single global header per stream and has timestamps so
storing xiph codecs in it should be possible but asf is patented and 
microsoft has already threatened individuals so we strongly urge you to 
avoid this container


Example and Disscussion of the matroska container
matroska supports storing 3 headers using a codec specific
format, which should be used for storing the 3 headers
Note, the above procedure to split one header into 3 works with the
vorbis-matroska specific format too


Example and Disscussion of the nut container
nut supports a single global header per stream so the 1<->3 merge/split
procedure above must be used, except that theres nothing special with 
storing xiph codecs in nut


Example and Disscussion of mpeg-ps / mpeg-ts container
These containers neither support a global header nor provide the neccessary
packet separation / framing, so storing xiph codecs in them is outside the 
scope of this appendix


Example and Disscussion of wav container
wav does not provide the neccessary packet separation / framing, so storing 
xiph codecs in it is outside the scope of this appendix


Example and Disscussion of the mov container
a single glbl atom shall be placed in the stsd atom in which the the global
header shall be stored