EveryZing Makes Everything Searchable

The Internet runs on text, which traditionally limits one’s ability to conduct a thorough search of multimedia files, yet the technology to do such thing, while still nascent, holds much promise. Indicative of this potential, EveryZing—a multimedia search and merchandising platform—announced earlier today a multitude of partnerships with Boston media firms Boston.com, 890 ESPN, boston.tv, and a variety of radio stations.  

Formerly the podcast search engine PodZinger, EveryZing spun out of BBN Technologies and now uses its expertise in speech-to-text technology to mine all multimedia files. Additionally, the company now has a business-to-business orientation, directly targeting media distributors instead of just the end users.  

"Search and text is the gulf stream of the Internet," says EveryZing CEO Tom Wilde. Bringing multimedia into text allows multimedia clips to be robustly indexed for search and—most importantly for enterprises—plugs multimedia into the search economy.  

"The value proposition is pretty clear when we approach these companies," Wilde continues. "They want fresh, unique users consuming the content, which is almost all ad supported. To have successful advertising business, you need inventory. To get inventory, you need users and consumption."  

Applied to audio mining, speech-to-text technology enables Web crawlers to discover previously unsearchable multimedia content and even deliver in-file navigation capabilities by searching within a file and diving directly to the section of the file where a relevant key word appears. For instance, an individual searching for New York Giants quarterback Eli Manning’s shoulder separation in a broadcast, instead of listening to the entire report, can leapfrog directly to the relevant news.  Companies—primarily major offline publishers like radio stations or newspaper sites—using a hosted search solution, according to Wilde, "are suddenly able to untap the value of these assets."    
 
Yet the technology is not perfect, and there are, as Wilde describes, two big hurdles in multimedia search. The most obvious hurdle is accuracy, but Wilde estimates that a comfortable zone resides north of 75 percent to 80 percent across a variety of content. "To do that," he says, "you need an extremely robust core technology and the processes to keep it fresh, namely updating dictionaries and language models and topic models."  EveryZing currently hires people to do that as a full-time job, another benefit to a hosted platform.  
 
The second problem is scalability. "What we find with many of our competitors," says Wilde, "is that they have robust technology, but it’s not built for the Web. Therefore it’s extremely expensive to scale to those levels."

Because the technology is still relatively new, its benefits are not immediately obvious to companies. Wilde cites a case involving one of the radio stations EveryZing hosts: "We’ve grown traffic from 14,000 page views per day to 200,000 page views per day using our solution," he says. "We know it works with our initial customers and we think it translates very nicely to our increasing customer base."    

SpeechTek Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues