Jennifer Zaino
SemanticWeb.com Contributor

As video becomes a bigger and bigger part of the content we consume - call it Video 3.0 - the challenge grows for publishers to tag that content and make it navigable, searchable, and monetizable. Semanticweb.com recently caught up with Alex Castro, CEO of Delve Networks, the semantically enabled and speech recognition savvy video search platform, who was fresh from the Digital Hollywood conference. Here's what he says some of the buzz is about, both at Digital Hollywood and throughout the industry, as it relates to next-generation online video content. On PC to TV video: At Digital Hollywood, Castro said there were a lot of discussions about how to get Internet video to consumers' TV sets.

"That's been something that's been kicked around but there's a bit more tangible traction around people solving that problem," he says. He pointed, as an example, to ClearLeap, which works with video content owners and cable, satellite and IPTV companies to offer a new model of video delivery on the Internet. "Essentially they tie Internet content companies into their system that works with set-top boxes," says Castro.

There's potential for Delve Networks to put its semantic video platform to work with ClearLeap. "Our idea with them is that the publishers using our system can get video piped to ClearLeap, to make sure it's in the right format so it looks good on TV. This is part of the recognition that video is becoming ubiquitous." Traction around solutions like ClearLeap's is good forDelve, Castro says, because it creates demand for publishers to want a video publishing platform.

On High-Definition video: Castro says there's been some surprise in the industry by consumer appetite for higher quality video on the Internet, which is leading to requirements to deliver high definition video online. The issue is that many of the consumers who want this don't have the high-speed connections required to deliver it seamlessly. That doesn't have an impact on the semantic capabilities Delve can bring to the video searching picture, but it does create challenges in terms of vendors being smarter about who they deliver video to, and how they do it.

"If there's a particular user with a terrible connection, and you are trying to give them high-definition video, that ends up in a terrible user experience, where you have rebuffering and jerky video and it's irritating," says Castro. "So, with a system like ours, we need to make multiple copies of a video at different bit rates and then try and determine, based on the quality of bandwidth for end users, which of those versions to serve, and maybe even adjust which bit rate you are using throughout as the video plays."

For example, for a long-form piece of video, it's possible that the connection degrades during the play time, especially for consumers using WiFi connections which ebb and flow in terms of quality, or for those using cable modems, where increasing neighborhood online traffic may disturb connection quality. "So our system has to be a lot smarter about what bit rate to use in any given time. We need an infrastructure on the back end to create many different versions of a piece of video, splice it up and have the intelligence that says, based on the quality of the connection, what to send down the pipe for a smooth video experience." Delve Networks is tackling that now and expects to provide a solution for this in Q1.

On Computer Vision. That's a topic that came up at Jupitermedia's recent Web 3.0 conference, he says. "Computer vision is an academic category that encompasses object recognition, object tracking, being able to search video with no audio," he says. It is the science of machines that see - for example, there might be a video where Barack Obama is speaking but he never says his name, yet the computer knows from the images that the video is Barack Obama talking. "There's excitement about what's possible there," says Castro. The industry sees opportunities for computer vision to play a role in everything from industrial robots controlling processes to organizing information of databases of images and image sequences.

Delve has done some exploring on that front but at this point doesn't have anything in the commercial pipeline to discuss. It's a very hard technical problem to tackle, he says, and not the first priority for its audience today.

"We focus on speech recognition and the semantic vision first, but computer vision will have its day," he says. It will be part of the whole intertwined web of extracting meaning from video, using a semantic layer to connect all the information collected from video, bringing speech recognition into the semantic layer, and applying intelligence to these processes to find the best results from end user perspectives. "Computer vision," he says, "is not magic bullet by itself."