acm-header
Sign In

Communications of the ACM

ACM TechNews

Boring Conversation? Let Your Computer Listen For You


View as: Print Mobile App Share:

Researchers are developing software that can make conversing with a computer more productive. Existing automatic speech recognition (ASR) technology is unreliable. "State-of-the-art ASR has an error rate of 30 to 35 percent, and that's just annoying," says University of Sheffield, UK's Simon Tucker.

The Massachusetts Institute of Technology's Alex Pentland says that even if ASR gets the words right, the results can be unsatisfactory because transcribing speech often makes for awkward reading. Tucker led a research team that developed Catchup, an intelligent ASR system that summarizes what has been said at a meeting. Catchup can identify the important words in an ASR transcript and edit out the unimportant ones. It measures the frequency of a word to calculate its importance and presents the results in audio form. The audio summary preserves some of the social signals embedded in speech, which could be lost in a simple transcription.

Meanwhile, Pentland is leading a research effort to develop Meeting Mediator, a device that measures how much time four people participating in an audio conference spend talking.

From New Scientist
View Full Article – May Require Free Registration

 

Abstracts Copyright © 2010 Information Inc., Bethesda, Maryland, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account