MIT Speech Product Helps Students Find Lectures

MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) unveiled Lecture Browser, a search engine that uses proprietary speech recognition technology to create transcripts of lectures and allow students to locate specific topics. 

Though the engine has been in development for four years and went live nearly half a year ago, CSAIL is only now publicizing its existence. "We were playing and toying around," says Regina Barzilay, an associate professor in electrical engineering and computer science. "You just need to make sure it works well before you advertise it." 

Barzilay, along with CSAIL principal research scientist James Glass, led a team of researchers and students to spearhead the project. Glass worked primarily on the speech recognition engine. while Barzilay focused on segmenting the transcripts.

"You take the transcript and divide it so when you retrieve it you do it topically," she explains. "We generate the title for each spot, just describe what each segment means, like a table of contents type of summary."   

So far, roughly 200 lectures are catalogued in the database. On the Web site, which requires Real Player, users search for a topic such as "black hole." They can focus their search by specifying their topic in a more specific category, like mathematics or interplanetary weaponry. Following a query, a list of all relevant lectures appears, highlighting points of interest within the audio feed. A video of the lecture plays while a transcript streams below. 

One of the initial problems with the speech technology was the need to specifically program some of the more esoteric terminology into the recognition engine. And even then, the transcripts generated are decipherable, but will occasionally miss accented phrases or words spoken too quickly. Still, the technology’s use is exciting.

"At MIT, a lot of lectures are recorded," says Barzilay. "And there are also some classes that are not only recorded but where people transcribe them by hand to make them presentable to students. It’s very expensive and time-consuming to do it. It’s a good place where you can use current technology to solve a real problem."

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues