Putting the Zing Into Multimedia Search: Q&A with EveryZing
One of the annoying parts of trying to find video and audio on the Internet is that you are completely dependent on descriptions attached to files. Miss the right keywords and you've probably missed the chance of finding what you're seeking. Even when you've found a clip, you sit through the whole thing looking for the moment you wanted. That's why EveryZing, Inc. is so interesting. The company is a spin-off from long-time tech hotbed and Department of Defense contractor BBN Technologies and has an interesting approach to making video and audio more readily accessible: automate transcript creation in real time and let people text search for the words in the content. It employs 30 people and runs the video search on about 120 web sites. We spoke to EveryZing CEO Tom Wilde.
BNET: Can you tell us a bit about the history of the company and technology? Tom Wilde: In 2001, the government thought that speech-to-text was a core piece of the national security technology infrastructure. The spec for the technology was to be speaker-independent, topic-independent, language-independent, and have tremendous scale requirements. If you're going to index millions of phone calls or hundreds thousands of hours of video per month, that's a web scaled problem. If the technology was commoditized, it would have bought off the shelf. They spent $50 million on research grants to BBN between 2001 and 2006 to deliver on this product specification. BBN successfully delivered that to the government, and then decided it would make sense to spin out everything. EveryZing was spun off from BBN two years ago. I came in as an outside CEO.
BNET: What is EveryZing doing with the technology? TW: The company has taken the BBN speech-to-text intellectual property and built what we call a media merchandising solution. We wrap video and audio with rich meta data and then we make that content discoverable and searchable. On the web you're used to being in control; you search and pick what you want. To date, video on the web has been another stream. Not only do I want to find a piece of video that's interesting to me, but I want to find the precise moment of the video. I want to forward the video as a clip to friends to include it on my blog.
BNET: How well protected is the company's IP?
TW: BBN has 19 issued and pending patents on this technology. The fact that it took $50 million this next generation capability means there's [also] a high wall around the technology. You have intellectual property protection and just a lot of research. Unless you have a piece of technology that can scale to web scale, maintain its accuracy, and operate on the web, you can't compete with us.
BNET: What about the technology itself? TW: The core technology is as good as we thought it is. It really has scaled nicely and delivered on the accuracy and throughput we hoped it would produce for web multimedia. It does it near real time. If we get a 3 minute video clip, it takes us about 3 minutes to process it. Out next closes competitor is about 4 or 5 to 1. That's the next tier down. If we get it streamed real time, we need a 15 second buffer.
The accuracy for a good piece of audio ranges from 80 to 90 percent. One key point is our approach is fundamentally different from out competitors. We built a cloud-based solution. Speech to text accuracy is driven by your ability to maintain your speech to text dictionary. All of our customers get the advantage of the continuous improvement of the increased accuracy.
BNET: You mentioned cloud-based, and that seems to becoming one of the buzzwords that is being applied to almost everything. Do you just mean that you host for other companies? TW: Cloud-based computing means unlimited capacity by leveraging remote computing power to deal with spikes in demand. We have our own hosted data center, but our technology has an agent architecture. I could set up machines on Amazon EC2 and IBM's service and basically provide unlimited capacity. I think there is a distinction. We can combine ours with theirs to provide 'unlimited' capacity.
BNET: How are you going to market? TW: We launched our first customers last fall. [Some of our big clients are] Boston.com, Fox Sports, Cox Radio, CBS Radio, and Fox News. Radio has been one of our really strong markets. We're part of Jott's work stream. We're in the market early. I think the customer reaction to the solution has been really strong, and our ability to get in front of the right customers has been terrific, but it's still something of an educational sale. People aren't used to searching for content this way.
BNET: This would seem good for such applications as electronic discovery and creating transcripts. Any thoughts on new products in the future? TW: We get plenty of requests for behind the firewall applications, and that's on the firewall. I would say sometime late next year we start to get after that.