Academics unite to debug speech systems

Von Michael Crawford

If you"ve found screaming at automated voice recognition systems pointless, things could be about to change.

A new research network established to iron out the bugs infesting automated telephony speech recognition systems is looking at ways to produce automatic speech recognition that can adapt automatically to a user"s state of mind - and ways for the systems to recognize and respond to agitation in the voice of the caller.

Named the Human Communication Science Network (HCSNet) and led by the University of Western Sydney and Macquarie University, the network aims to link scientific, engineering and IT professionals with psychologists, creative artists, musicians and linguistics experts.

HCSNet convenor, Professor Robert Dale said the research is not about creating an artificial human but a more "human-like" communication with machines.

"We will be drawing on the subtleties and nuances of human-to-human communication to improve human-machine interactions," Dale said.

"The new group is designed to improve existing technologies and encourage new concepts and ideas that will make our lives easier."

Chair of the network"s training and development committee, Associate Professor Kate Stevens said it is a long-term, sustainable project.

"Not only will we provide the vehicle for unique collaboration, but we will also train and mentor younger, less experienced researchers so they can continue the quest for more effective communication tools in years to come," Stevens said.

Steering committee chair Professor Denis Burnham said the project is about finding a new approach to an old problem, and will be a platform for new ideas and creativity.

"This network will break the conventional interdisciplinary research links and forge new collaborations across fields as diverse as psychology, computing, linguistics and language development, acoustic science and auditory neuroscience," Burnham said.

"We believe their combined expertise will be the perfect crucible for new ideas and creativity."

HSCNet currently involves 17 Australian universities, research institutes in Japan, the UK, US, Canada, Asia and Europe and many Australian private and public companies.

The HCSNet is funded through an Australian Research Council grant of A$2 million over five years, and will be launched today at the MARCS Auditory Laboratories at the University of Western Sydney"s Bankstown campus.

Other potential research areas include machine translations in booking systems, more efficient methods of data mining and information retrieval and combining research in IT and auditory science.