Google has been indexing Flash since 2004 but the underlying technology that powers millions of front pages and zillions of banner ads has defied the kind of interrogation that makes it truly searchable, until very recently. At the beginning of July, Adobe released what it calls Searchable SWF technology but although this is a step in the right direction
… only text and links will be searchable. Graphics and video, including FLV files, still won’t be able to be indexed properly, and there’s no capability to search and index metadata embedded in Flash files (even though, Adobe says, SWF and FLV files have metadata fields) or to allow people to link to specific content within a Flash file in order to make search results more relevant. Also, when someone clicks on a search result, they’ll be taken to the beginning of a Flash file and will have to navigate their way to the content they are seeking.
The reality is that while Text is still King for search engines, more and more eyes and ears are engaged with media online other than the typed. In the past three years, by far the largest increase in traffic across the net as a whole has been in streaming video, jumping from 12 % in 2006 to 22% in 2007. Video is predicted to account for as much as 90% of total in five years time, and so current search technology will have to evolve to deal with this situation, or else it will lose its foothold to newcomers.
There are several emerging technologies which would seem to have the potential for searching within video and audio. Blinkx is probably the market leader - it is
…based on technology that was conceived at Cambridge University, enhanced by $150M in R&D over 12 years, and is now protected by 111 patents.Unlike other multimedia search engines that attempt to re-purpose technology built for the Text Web, blinkx uses a unique combination of patented conceptual search, speech recognition and video analysis software to efficiently, automatically and accurately find and qualify online video. Today, blinkx is the world’s largest single index of rich media content on the Web, delivering more content from a broader range of sources than either Google or Yahoo!
Pluggd offers something similar, as does Everyzing (once called Podzinger). These should allow the user to pinpoint key phrases and words within a speech-based soundtrack, find relevant visual objects within video, etc.
The biggest problem is that audio and video has to be analysed before it can be useful, and because of the sheer volume of data involved, this process does not lend itself to the modern, impatient world. My experiences with online speech recognition have been that it works best with well-modulated, carefully constructed mid-Atlantic accents, and the speech-to-text translations can offer up words pronounced “bacon” as “beer can” if spoken in a Geordie or other similarly inflected version of an English accent.
With online video not always being of the best technical quality (and people not seeming to mind that too much either) video search, which works well in lab conditions where objects are clearly delineated and a can of beer looks like a beer can, might also fall into the same trap, so that only the most obvious, simple and clearly produced videos will be properly searchable. How boring would it be if ads (for example) in order to be internet-worthy could only be made in primary colours with slow, unexeptional speech patterns? What will Blinkx make of a Guiness ad?
*UPDATE: July 30th 2008: Blinkx Launches Web TV Search (Guardian)*
BlogDigger on the other hand, like many such smaller search engines, searches via RSS, the ubiquitous technology that underpins blogs, podcasts, and allows anonymous subscription to content. It’s quick, light, and so long as the producers do their stuff, it’s accurate. For the moment, and for the foreseeable future, show notes and accompanying text remain the necessary accompaniment to all forms of internet media, if you want to create a context in which it can be found for the reasons you want it to be found.
Sphere: Related Content
