- From: Conrad Parker <conrad@metadecks.org>
- Date: Thu, 26 Mar 2009 10:46:51 +0900
- To: Jack Jansen <Jack.Jansen@cwi.nl>
- Cc: Media Fragment <public-media-fragment@w3.org>
2009/3/26 Jack Jansen <Jack.Jansen@cwi.nl>: > I spent most of the evening battling ffmpeg to get it to do what I want, > with only limited success. I can get it to leave the data intact (by using > the "copy" codecs), but the command line utility will always muck up the > timestamps. This is solvable, but it would involve creating a new > application (ffclip?) to do exactly what I want. > But in the process I have learned something interesting. ("interesting" is > used here in the same sense as in the Chinese curse "may you live in > interesting times!":-). > If our original media is well-behaved and has both audio and video starting > at exactly the same timestamp, and we disallow transcoding we lose this > property. > Our audio packets will, in general, not have the same timestamps as our > video packets, so the resulting fragment has audio and video starting at > different times. This is not only a burden on the playback implementation > (which isn't all that much of a burden: it'll probably have audio resync > code anyway), but it also means that in our HTTP extension we cannot use a > single timestamp to denote the start (or end) of our fragment: the audio and > video stream will have slightly different start timestamps. yes, that's why we developed Ogg Skeleton, which ffmpeg does not yet support. http://wiki.xiph.org/OggSkeleton The problem generalizes to different codecs. For Ogg, we have already solved the start/end timestamps and the HTTP details in the Annodex project. Conrad.
Received on Thursday, 26 March 2009 01:47:31 UTC