Intel goes open source with new speech-recognition software

Posted on 29 Apr 2003 at 11:36

Intel researchers have been working on a new form of speech recognition software.

As well as processing sounds produced by a person, the system will visually process the face for further information. It will, in Intel's words, 'read lips'.

The simple premise is that this is the way humans better understand conversations. 'Intel wants to develop technology that allows computers to naturally interact with the world the way humans do,' said Justin R. Rattner, director of Intel's Microprocessor Research Labs. Human recognition is seldom based on a single type of information. We make decisions by combining information from a variety of sources.'

Specifically, Intel's Audio Visual Speech Recognition (AVSR) software is combined with face detection algorithms from Intel's OpenCV computer vision library. These detect a speaker's face and mouth movement for more accurate speech processing.

Intel also makes the point that traditional speech-recognition algorithms work best when bad noise is eliminated or a well-tuned microphone is used. This is not always a realistic requirement and the new approach helps surmount the problems. As long as the lighting is of a sufficient level, of course.

The work has been released under an open-source license, to encourage third parties to pick up the software for further research and development. Very altruistic, but of course speech processing is an area of computing yet to come to full fruition. When it does, processing power will be important. Which is where Intel comes in.

You can find more info about Intel's OpenCV 'toolbox', to which the AVSR software has now been added, at www.intel.com/research/mrl/research/opencv/index.htm.

Author: Alun Williams