OMwiki:Tech

Revision as of 01:49, 25 November 2009

Known issues:

• Find a way to add style="width:352px;" to embed code in 352×240 streams and fix aspect ratio for 352×240 thumbnails when in category view.

• Video playback might take ~6 seconds to start for timecodes towards the end of the meeting (pending seeking support in oggz-chop). Also, Google Chrome won't start playback if oggz-chop is used.

• Edit MediaWiki:Common.js to download animated gifs only when 'show' is clicked to avoid long page loads.

• Speaker labels work for annotations but not transcripts. Also need to figure out AJAX auto-complete magic.

• Search works but has a few issues (punctuation stripped, 'play inline' plays audio although video is paused, search-by-date not yet enabled for all videos)

Scripts needed:

• Retool download_from_archive_org.php to work with different source videos and meta tags from the Internet Archive, using this wikipage as a working draft.
• Draft a script to import MetaVidWiki clips from apps that can produce .srt or .cmml captions (e.g., Gnome Subtitles)
• Rework modcopy to avoid mmap calls on JFFS2 filesystems (e.g., the OLPC-XO). Ticket

Things that make life difficult:

• ffmpeg/mplayer seem to handle duplicate Theora frames poorly, breaking A/V sync and thus limiting video reuse.

• Cinelerra-CV is buggy and somewhat dated. Document rendering options on other video editors (e.g., XiphQT) and help develop Lumiera.

• Fix or forgo non-compliment DV hardware. Example ticket

• USB headsets crash spontaneously (error message usb_set_interface failed, at least in Debian Squeeze).

New features

• Ideas on MetaVid-l

• Add a Vorbis-only option on video streams for low-bandwidth connections.

• Turn on OpenID logins. Need to figure out how to link non-PEAR libraries to non-recompilable PHP.

• Replace animated GIFs with low-FPS, enlargeable, jump-to-timecode Theora. Alternatively, use Pad.ma-style mouse-overs (see $(imagePoster).load(function() {...})). The thumbnails should cycle through the meeting in large time chunks when non-mouseover'd, and then display clickable frames according the how far the mouse is from the left edge when mouseover'd. Include ability to pause all animations and step-through lock-step upon mouse clicks.

• Add YUV4MPEG support in Blender to enable direct output to ffmpeg2theora (email thread), building upon File:Henningsen 7oct2009 blender-svn.tar.bz2 Also build-in yuv4ogg support as per xiphmont's lead.

• Squash bugs (esp., video non-playback) in mwEmbed's libSequencer.

• Identify a way to output time ranges generated from multi-speaker audio files, with each time range corresponding to the duration of how long a person spoke for (CMU Sphinx?). Hopefully add speaker identification based on the vocal profile, and maybe even facial recognition and emotional categorization (e.g., laughter).

• Figure out a content delivery network and targeted areas.

Dreamlist

• Build open-firmware cameras that record directly to Ogg (2-pass Theora plus Vorbis); also include simplified on-camera editing. Elphel has completely-open network cameras.
• Accellerated Theora encoding using hardware acceleration (TI C64x+ DSP or an Elphel card connected via PCI?)
• Have those same cameras set to blur-out individuals broadcasting a 'do not record my voice, person, or both' flag via iPhones, RFID, or colored stickers as mentioned in Yochai Benkler's keynote at OVC 2009.
• Have individuals appear as virtual avatars in-video (i.e., Second Life-esque), according to user preference.

Communicate:

• Join and post to MetaVid-l
• Join #metavid and #mediawiki. No IRC? Click here to join via a webclient.
• Have your work recognized on Sunlight Labs: Projects by contributing code.

Non-technical development

• Please see OMwiki:Volunteer