No, it’s not a flying car, but it’s pretty damn cool. MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) released a new lecture search tool that facilitates keyword searches of transcripts from about 200 recorded lectures.
A Lecture Browser, by CSAIL at MIT
The following text comes from a transcript of a video made by MIT:
conventional search engines are all text based and that’s been very effective it’s great there’s all sorts of text on the web but it doesn’t do anything for audio and video materials and this type of data is just exploding these days it’s becoming easier than ever to create to store to disseminate these kinds of data just look at the explosion of podcasting for example this project is all about letting people search inside video material of recorded lectures to find particular snippets that they’re interested in … what the demonstration that users are going to be able to try out today is sort of demonstrating the current status of some of these different technologies here we have the lecture browser web interface … you can see at the top that it is much like any conventional search engine … i can type in this text box topic keywords i’m interested in or pick a category … so just for example why don’t i pick the arts and humanities categories and click search … and you can see when you look at the results there are three videos in this category right now … at the bottom here for example we have a lecture by sendak who is the author of where the wild things are … the grey rectangle represents the entire lecture … and the computer has taken the lecture and automatically divided it into these segments … if i wanted to i could actually click below where it says text and get some examples of what was actually said in this particular segment … if you’re interested in any particular segment you can just click on the play button here so why don’t i do that … you have the video screen … and down below it we have the automatically
passionate about in my research is speech speech is natural for people people communicate with language and speech is the main modality they use to interact with each other my whole research is devoted around developing or advancing speech and language technology to help people both communicate with a computer by speaking naturally and also as in this project … annotating providing structure to speech content to help people get better access to this kind of information
Clicking on any keyword in that transcript will instantly jump you to the specific point in the video where the text occurred. The transcripts are generated with advanced speech recognition software, augmented by researchers feeding more technical terms into the system. The system gets four out of every five words correct, with the errors mainly on words not critical to the lecture topic. Researchers Regina Barzilay and James Glass—who presented their project at the Interspeech 2007 conference in Antwerp, Belgium, in August—hope to incorporate Wikipedia-like tools to allow the community of browser users to make corrections.
The project, originally funded by Microsoft through the iCampus program, is now funded by the National Science Foundation. The video display requires the RealPlayer plug-in.
