Tuesday February 27, 2007

Searching digital video

An interesting article appeared in the NYT business section on Sunday, http://www.nytimes.com/2007/02/25/business/yourmoney/25slip.html?_r=1&oref=slogin
about a company called Blinkx. Blinkx is trying to make digital video searchable and the article discusses their methodology and why it appears to be more successful than Google Video. GV searches metadata associated with video, whereas Blinkx is trying to automatically transcribe speech in videos (an extremely nontrivial problem) and then search that speech rendered into text.

I believe this merits digital library attention because when we discuss institutional repositories, there is always a focus on papers or data, but  video doesn't seem as prominent. Yet video may be an important part of what IRs or future university archives end up holding. For instance, last Friday night Blanton Godfrey, the Dean of the College of Textiles, appeared on Watch North Carolina People with Bill Friday. His conversation was wide ranging. I'm trying to add a DVD/video of his appearance to the Textiles Library media collection. But if someone wanted to know if he spoke about a particular subject in his roughly half hour interview, there's no easy way to find out, without searching a transcript or watching the entire show.

This type of content may become more common in the future, especially for prominent individuals. A few weeks ago a number of us had an ongoing discussion about a concept called 'lifelogging' where a person tries to keep track of their entire life digitally (or entire digital life; the two are as yet distinguishable, the second being easier and more common). If video becomes something we collect in earnest, both for repositories and for regular collections, then being able to search that content in a meaningful way (i.e. not GV) becomes important.

It is, however, worth noting the limitations of what Blinkx is trying to do. The last paragraph in the article emphasizes that Blinkx is searching the sounds (or rather transcribed text) of a video and cannot do anything about the images in a video. Because so many videos (unlike the example I mentioned above) may not have much speech to go by, there is still no good way to search them (Blinkx for all intents and purposes is just creating much more textual metadata for a video). Imagine, for instance, trying to search Koyaanisqatsi to find out if there are depictions of certain places/jobs/landmarks, natural events, etc.

Try Blinkx out at http://blinkx.com/