theora dev - Sep 2007 - AW: How to get the duration of a file or track?

Hello,
> (1) Split the content into chain segments 
> (2) extract the first and last data packet from each segment, for all
tracks
> (3) recover / calculate the granulepos values for the first and last data
packets (*)
> (4) convert the granulepos values from the first and last data packets into
time values
> (5) take the largest difference between start and end time as the total
chain segment time
> (6) sum the chain segment times to create the total content length
those steps to determine the duration of ogg/theora streams are really annoying
because all ogg pages have to be parsed through the whole file (if I got it
right?).
This takes an extra amount of time but there are applications where you simply
can't afford any further delays so the the video/audio has to start
immediately. Not to mention playback from CD-ROM in combination with seeking
abilities.

Why there's no simple duration entry in the bos page of every stream set by
the ogg encoder specified? Ok, in case of a real streams which doesn't have
a defined length (live recording) the entry is just left empty. But in all the
other cases you convert an already existing file container (like AVI or MOV
where the length is known) to ogg.

Currently I would tend to write an ogg/theora encoder which adds an duration
value to the comment block so its easy ascertainable by the decoder.

Cheers,
Robin.

On 26/09/2007, Robin Siegemund <r.siegemund@digitalpublishing.de>
wrote:
>
> Hello,
>
> > (1) Split the content into chain segments
> > (2) extract the first and last data packet from each segment, for all
tracks
> > (3) recover / calculate the granulepos values for the first and last
data packets (*)
> > (4) convert the granulepos values from the first and last data packets
into time values
> > (5) take the largest difference between start and end time as the
total chain segment time
> > (6) sum the chain segment times to create the total content length
>
> those steps to determine the duration of ogg/theora streams are really
annoying
Yes these steps are annoying. Worse, they are inexact as the
granulepos of the first data page in each bitstream refers to the
ending time of the last packet on that page. To get an exact duration,
the duration of these first packets needs to be calculated by
examining the data and added to the figure obtained in step 5 above.

Alternatively, the starting granulepos for each track is given in the
"basegranule" fields of Ogg Skeleton.
> because all ogg pages have to be parsed through the whole file (if I got it
right?).
Not all pages need to be parsed; only the BOS and EOS pages need to be
found for each chain, and these can be found by binary search.
> This takes an extra amount of time but there are applications where you
simply can't afford any further delays so the the video/audio has to start
immediately. Not to mention playback from CD-ROM in combination with seeking
abilities.
>
> Why there's no simple duration entry in the bos page of every stream
set by the ogg encoder specified? Ok, in case of a real streams which
doesn't have a defined length (live recording) the entry is just left empty.
But in all the other cases you convert an already existing file container (like
AVI or MOV where the length is known) to ogg.
>
> Currently I would tend to write an ogg/theora encoder which adds an
duration value to the comment block so its easy ascertainable by the decoder.
Yes, or a Content-Duration header in Ogg Skeleton, similar to what
oggz-info prints out (or similar to these HTTP headers:
http://trac.annodex.net/wiki/HttpHeaders )

cheers,

Conrad.