ogg dev - May 2008 - How Ogg mappings translate into the codecs parameter in Ogg media types

Hello list,

We have recently introduced the codecs (optional) parameter on the Ogg
media types for easier identification of encapsulated codecs in
conditions were it's not feasable to examine the streams directly.

I received a comment from the IESG that we should have some sort of
list to describe how Ogg mapping strings translate to codec parameter
strings, so I went ahead and compiled a table with all known Ogg
codecs, and I would like feedback in case I inserted an error
somewhere.

The draft is located in SVN[1], but for your convenience I'm posting it
here:

 	        Codec Identifier           | Codecs Parameter
 	       -----------------------------------------------------------
 	        char[8]: '\x01vorbis'      | vorbis
 	        char[8]: '\x80theora'      | theora
 	        char[8]: 'Speex   '        | speex
 	        char[8]: '\177FLAC'        | flac
 	        char[8]: 'BBCD\0'          | dirac
 	        char[8]: 'OggMIDI\0'       | midi
 	        char[8]: 'CMML\0\0\0\0'    | cmml
 	        char[8]: 'PCM     '        | pcm
 	        char[8]: 'CELT    '        | celt
 	        char[8]: 'kate\0\0\0\0'    | kate

Is everything correct?  Did I miss a codec?

-Ivo

[1] https://trac.xiph.org/browser/trunk/standards/draft-goncalves-rfc3534bis.txt

2008/5/24 Ivo Emanuel Gon?alves <justivo at
gmail.com>:
>
> I received a comment from the IESG that we should have some sort of
> list to describe how Ogg mapping strings translate to codec parameter
> strings, so I went ahead and compiled a table with all known Ogg
> codecs, and I would like feedback in case I inserted an error
> somewhere.
>
> The draft is located in SVN[1], but for your convenience I'm posting it
here:
>
>                Codec Identifier           | Codecs Parameter
>               -----------------------------------------------------------
>                char[8]: '\x01vorbis'      | vorbis
>                char[8]: '\x80theora'      | theora
>                char[8]: 'Speex   '        | speex
>                char[8]: '\177FLAC'        | flac
>                char[8]: 'BBCD\0'          | dirac
>                char[8]: 'OggMIDI\0'       | midi
>                char[8]: 'CMML\0\0\0\0'    | cmml
>                char[8]: 'PCM     '        | pcm
>                char[8]: 'CELT    '        | celt
>                char[8]: 'kate\0\0\0\0'    | kate
>
> Is everything correct?  Did I miss a codec?
The magic is not always char[8] -- it's an arbitrary number of
characters, and as far as I know there is no official limit. So it's 7
for Vorbis and Theora, 5 for FLAC, and so on. eg. immediately after
'\177FLAC' is a version field, which is not part of the codec ident
string but is within the first 8 bytes.

What you've listed for CMML is correct, and I'm pretty sure the
strings are correct for Vorbis, Theora, Speex, and PCM.

Also I don't know if the Dirac, MIDI, CELT or Kate mappings are yet
finalized?

cheers,

Conrad.

On 5/24/08, Conrad Parker <conrad at metadecks.org>
wrote:
> The magic is not always char[8] -- it's an arbitrary number of
> characters, and as far as I know there is no official limit.
I was under the impression that 8 was the limit.  Thus the extra white
space JM adds on Speex and CELT magics.  I may be wrong, though.
> So it's 7 for Vorbis and Theora, 5 for FLAC
So char[7]: '\x01vorbis', char[7]: '\x80theora' and char[5]:
'\177FLAC'?
> immediately after '\177FLAC' is a version field, which is not part
of the
> codec ident string but is within the first 8 bytes.
I don't think that interferes with anything.  We are talking about the
essential part needed to be detected by applications.
> What you've listed for CMML is correct, and I'm pretty sure the
> strings are correct for Vorbis, Theora, Speex, and PCM.
Good, good.
> Also I don't know if the Dirac, MIDI, CELT or Kate mappings are yet
> finalized?
Kate is AFAIK finalized; that's the impression I get from talks with
ogg.k.ogg.k.

Nobody's touched MIDI in years, so that's probably how it will stay.

Considering the similarity between the CELT and Speex strings (the
extra white space thingy) I don't think JM's changing it.

Dirac is the only one I'm not sure, since this one here's the recent
string.  The magic used to be different, and I'm not sure if whoever's
working on Dirac may have a change of ideas.

Anyway, the essential for the document is that no errors can be
present.  Changes and newer codecs may be added later on an update to
the document.

Thanks.

-Ivo

>           char[8]: 'kate\0\0\0\0'    | kate
This is incorrect, should be:

            char[9]: '\x80kate\0\0\0\0'    | kate

As Theora does, each packet has a first byte for the packet type (MSB
set for headers).
The padding with 4 zeros rather than 3 is due to, er, historical reasons :)

> Also I don't know if the Dirac, MIDI, CELT or Kate mappings are yet
> finalized?
The Kate one is - while the bitstream is still evolving (keeping backward
compatiblity though), the mapping is not. It helps that Kate's designed as
a string of packets, so the mapping's very simple.

On 5/26/08, ogg.k.ogg.k at googlemail.com <ogg.k.ogg.k at googlemail.com>
wrote:
>>           char[8]: 'kate\0\0\0\0'    | kate
>
> This is incorrect, should be:
>
>             char[9]: '\x80kate\0\0\0\0'    | kate
The last one is how it's listed on ogginfo code, but the first one
(char[8]) is how it's described in the wiki page.  I thought the
latter would be more correct.  If not, you should revise the wiki
page.

Thanks for the help, everyone.  I'm committing changes based on this
thread to the draft.

-Ivo