Alex Laats, PodZinger
Click here to view video and read post from Beet.TV
[Start of Audio]
Alex Laats: When it comes to a search for audio and video content, the key is that you actually have to have access into the words that are spoken inside of the video or audio content. In the absence of that, all you’re searching are titles and maybe a summary, information that the author of the content created, such as show notes. Or user generated tags.
Andy Plesser: Alex, tell us a little bit about PodZinger and what you guys do up there and how the technology works.
Alex Laats: We operate a farm of servers that run speech to text technology that come from PodZinger’s parent company, BDN. That farm of servers is – basically ingests audio and video. It runs it through by, in essence, watching it once, and takes every word spoken in an audio or video segment and tags it with the identification of that word. We call it a rough transcript, or what I really call it is a text index. So think of it as – much as sites – web ______ _______ sites are very much about tagging media in order to be able to create additional value and access to that media. This is about trying to tag every word with the identity of that word and the location of that word as it occurs in the video.
And so, of course, much like humans, more than humans, computers are imperfect. They’re just not perfect, so they’re not going to get a perfect transcript. So – however, they create so much more metadata about video and audio so that a user can actually find content that’s relevant to them that would otherwise be black space to Google or Yahoo.
Andy Plesser: I understood that you folks really index podcasts or files that’re sent to you by RSS. Now you’re doing user-generated content and you’re indexing and organizing YouTube files? Can you explain that a little bit?
Alex Laats: We literally take every file that’s produced on YouTube, and we’ve done this since the beginning of December so now we’re at somewhere around 1-1/2 million videos in just over three months. Okay, so it’s massive amounts of video that’s being created, and we run that through our farm of servers, and we take all the metadata that’s associated with the content on the YouTube page and then add a whole lot more starting with things like is it music or is it not music? Then we say okay, and how much of it is rich in word – kind of word density? For that content, we run it through our speech detection engine and create a rich text index of that piece of media.
So what that does then is it creates a much richer set of metadata and tags associated with each video. On top of that, we add topic classification. So what we do is – we think we help make that – turn what is kind of a hard to access, hard to consume, hard to advertise collection – massive collection of media, and we turn that into an easy to consume, easy to search, easy to advertise collection of media.
[End of Audio]
video search, 
YouTube, 
PodZinger, 
Alex Laats
Comments