Ideas to tackle online media captioning

For years I’ve thought about how best to deal with captioning for webcasts and now podcasts in a cost-effective and scalable way. I believe that the time is right to tackle this with a combination of Wikis and a slick transcribing interface modeled on dotSUB, a collaborative subtitling program.

Beyond the critical need to make webcasts and podcasts accessible to hearing-impaired there are additional benefits:

• Written transcript often easier to peruse and annotate
• Keyword/phrase search results synchronized to exact moments in the media
• Same language subtitling helps literacy and/or learning scientific terminology
• Learning disabled accessibility
• English as a second language learning
• Transcripts can be translated to extend the media to a worldwide audience

Challenge #1: Producing the transcript
Hiring a professional transcriber is about $90/hour. The CastingWords service leverages MTurk to bring the price down to about $.42/minute, or $25/hour. This has proven cost-effective and accurate enough for folks like Jon Udell to go for it. I’ve heard some reports about inaccuracies that lead to further clean-up.

I believe that volunteers are inspired to transcribe educational media. This is inspired by the work of OOPS (Opensource Opencourseware Prototyping System). Based in Taiwan, the volunteers translate Creative Commons licensed work primarily from MIT’s OpenCourseWare in order to bring the knowledge to Taiwan and to learn English.

How it works, from their site:

As a volunteer-based project, we ask you to adopt the file(s) you wish to transcribe… Once you have located the lecture(s) you wish to transcribe, simply click on the “Edit” link on the top right-hand side of the page… Once you have finished transcribing, come back to this site and copy and paste the text into a placeholder that will have been setup for you.

As you can see from the list of courses, however, it may take awhile for all of the courses to be adopted and finalized. OOPS also does not seem to use a Wiki interface for their work. At the end of the day, CastingWords is probably the most direct route but I’m still fascinated by the Wiki.

I’ve looked into Wikia but am amazed to not find one for transcribing (and I don’t want to kick it off!). Please let me know if one exists. Perhaps we should simply set up our own private wiki for this. Ideas welcome.

Challenge #2: Synchronizing the media
Transcript in hand, the next step is to produce timecodes for captioning. At Berkeley we love Automatic Sync Technologies for this. Their software automatically takes a transcript and the media (an MP3 of the video is preferred) and quickly produces timecoded transcripts in a variety of formats. This still costs  roughly $60 per hour of content, though less if hours are bought in bulk.

Again, I believe that with the right interface volunteers can chunk the transcripts, too.

Transcribing interface
Ideally, we would have a UI layered on top for people to transcribe while viewing or listening to the media. The UI would have quickkeys for controlling the playback (start, stop, pause). The input fields would be pre-chunked to facilitate the timecoding.

This UI was inspired by dotSUB that I first saw demoed during Andrew Baron’s keynote at the Podcast Expo. dotSUB is still in beta, and their focus is collaborative translation of existing timecoded transcripts

According to dotSub’s film submission page, they require 1) the video to be uploaded; 2) the timecoded transcript.

Regarding #1, frankly I’m not keen on sending my videos to yet another 3rd-party. Furthermore, there is a Filmmaker Agreement which presents an instant hurdle.

Many of our videos are already up on Google Video, as are those of many others including YouTube, Blip.tv, etc. All of these services provide HTML snippets to embed videos into a webpage. dotSUB’s business model may include making the subtitled videos available as hi-res and/or portable distribution (DVD, etc.). If dotSUB accepted 3rd-party embeds, they’d have a lot more people using their system. I wonder if there is a technical issue with syncing up timecodes with videos being served from an external source.

Regarding #2, I briefly chatted with them and there is a possibility for opening up the interface for pure transcribing.

Exporting timecoded transcripts
Finally, I would love to be able to export the timecoded transcripts. I could then submit them alongside my Google Videos for close captioning, an important feature that they recently announced. iTunes U can now search upon keywords, but I don’t believe the results bring you to those moments in the media. Of course we’d like to add these to our local webcast.berkeley site where we do have caption search implemented.

Summary
Adopt-a-podcast feature + Wiki for collaborative transcribing and finalization + dotSub-like interface for a useful transcribing interface that also produces timecodes. It would be an interesting experiment, with the first step to set up a Wiki.