EveryZing Search-Optimizes Video with Speech-to-Text - Broadcasting & Cable

EveryZing Search-Optimizes Video with Speech-to-Text

Multimedia search specialist leverages BBN Technologies research.
Author:
Publish date:

In the six months since EveryZing began marketing its multimedia-search and search-optimization technology, CEO Tom Wilde said he has seen a shift in attitudes from potential customers such as television news executives.

The conversations started out "highly educational" as he tried to convince them why they needed what his firm is selling, but "now it has become something that's on their shopping list," he said.

For any organization trying to turn broadcast programming into Web content that will attract viewers (and the advertisers who want to reach them), one challenge is getting the attention of search engines that were built primarily to index text, not video or audio clips.

When Google or another search engine adds a Web page to its index, it starts by indexing every word on the page and analyzing contextual clues such as the content of headlines and the text associated with links. So for content that is fundamentally nontextual, there is simply less to work with -- often no more than a headline, a short description and a few manually entered tags. Often, that's not enough, particularly if the content is being slapped online in a hurry.

"What we tend to see are generic headlines and only moderate tagging. The title of a video might be, 'Daily News Roundup for July 30, 2008' -- that would be a very typical title for a clip on a TV news site," Wilde said. That tends to put TV news stories at a disadvantage compared with newspaper and wire-service stories that are all text.

EveryZing attacks this problem with speech-to-text software that automates the generation of program transcripts, which a search engine can then index.

The two-year-old company is a spinoff from BBN Technologies, the Boston-based information-systems firm known for its work on government contracts such as the launch of ARPANET, an early version of the military and academic network that became the Internet.

The automated-transcription technology, originally developed for government applications such as intelligence gathering, represents about a $100 million investment, Wilde said, and is significantly better than anything else on the market.

Products based on the technology include ezSEO, which helps Web-site operators make their video content more visible to search engines, and ezSEARCH, for searching individual Web sites. Customers include broadcast organizations such as Fox Sports and Boston's WCVB-TV, which uses the technology on its thebostonchannel.com Web site. Wilde said WCVB owner Hearst-Argyle Television is testing the technology there with an eye toward broader deployment if it proves successful.

EveryZing's transcription technology actually only attacks one aspect of the search-engine-optimization problem, given that search engines also look at how many other Web sites link to a particular Web page (or video) as an important clue to how to rank it within a set of search results. That's part of the reason why search engines already include some videos in their results without necessarily having access to a transcript.

But popularity-based search rankings, such as Google's PageRank, still depend on viewers being able to find the videos, Wilde pointed out. "If the content is not published in the first place in a way that it can be discovered, nobody's ever going to look at it," he said.

Related