[theora] Indexing Ogg files for faster seeking

[Gregory Maxwell]

On Fri, Jan 22, 2010 at 7:14 PM, Bernard Jungen
<bjung50169 at euphonynet.be> wrote:
> However, for Vorbis, I see the time stamp deltas are quite random. An unfortunate
> (in this case) property of Vorbis I suppose? The probabilities distribution
> of deltas, when sorted, seems to be exponential as can be seen in the
> following table, and an entropy coder may be able to work well on it:

Sort of the opposite—  Every Vorbis frame is seekable. To avoid having
a zillion index points it generates index entries only every 2000ms
and only if 64kb away from the last. Mix in the vorbis frames not
being an integer number of milliseconds long, there being multiple
frame sizes, and the indexer generating an index to the second vorbis
frame on a page .... and it ends up being quite noisy.

With some changes to the indexer the vorbis timestamps could be much
more predictable... but any compressor should also handle cases like
this, so I suppose its an okay example.