The web, especially in conjunction with Java, offers the potential of an unprecedented degree of platform independence, and consequent enormous reduction in costs to distribute, port and maintain software, data resources, and updates. Speech-enabled web interactions vastly increase accessibility to the web, including access from the nearest telephone (which may be in your pocket), and access by disabled populations, or whenever typing and mouse interactions are inconvenient.
Our first demonstration of speech-enabled web-access is based on the Air Travel Information System (ATIS). This system, based on speech recognition and natural language understanding technology, is capable of responding to spoken air travel queries (a frozen and outdated database of flight information is used in this demonstration).
For the first time, human language and Java technologies have been fully integrated. A central server, built around SRI's Open Agent Architecture (OAA) combines numerous modules that provide the required functionality: speech recognition, natural language understanding, database access, telephone control, and so on. A lightweight client written entirely in Java provides the user interface. The client can be run on any Java compatible web browser. Since there is yet no audio input standard API for Java, we have implemented the first version of the system using a telephone connection that provides audio input and output to the user.
This demonstration is now available on the web at (http://www.speech.sri.com/demo/atis.html).
The Speech Technology and Research (STAR) Laboratory at SRI is a world leader in speech technology. SRI's DECIPHER technology is capable of recognizing natural, continuous speech without requiring the user to train the system in advance (i.e., speaker-independent recognition). The DECIPHER system exhibits leading performance in government-administered tests and can be distinguished from the few other leading-edge speech recognition technologies by its detailed modeling of variations in pronunciation and its robustness to background noise and channel distortion. These features make the DECIPHER system accurate in recognizing spontaneous speech of different dialects and less dependent on the idiosyncracies of different microphones and acoustic environments.
SRI International is an independent, nonprofit corporation chartered by the State of California, performing a broad spectrum of problem-oriented research under contract to government, business, and industry. SRI serves clients in all parts of the United States and throughout the world.
For more information on speech-enabled web access contact:
Leonardo Neumeyer Speech Technology and Research Laboratory SRI International 333 Ravenswood Avenue Menlo Park, CA 94025-3493 email: leo@speech.sri.com
Decipher, Open Agent Architecture, and OAA are trademarks of SRI International. Java is a trademark of Sun Microsystems, Inc. All other product or service names mentioned herein are trademarks of their respective owners.