Spoken User Interfaces
Bob Carpenter SpeechWorks International,
Seminar on People, Computers, and Design
Stanford University April 7, 2000
My talk provides a practical, step-by-step overview of the issues surrounding spoken user interfaces. Concentrating on telephony call center applications with natural spoken language interfaces, I will cover the entire application lifecycle: design and prototyping, coding, back-end and platform integration, deployment, tuning and testing.
I will detail the process we use at SpeechWorks, beginning with customer-centered user-interface design and prototyping, which leads to a detailed specification. I will then discuss the issues involved in coding the specification on a development environment, of which we use several telephony specific platforms and plain old C++ code. I will then focus on how interfaces are ramped up during deployment, during which time tuning and testing is fairly intense.
I will conclude with an overview of some of the future directions in spoken user interfaces, focusing flexibility and naturalness.
I will cosnider the design of a dialogue system to accomodate various kind of user profiles, from naive first-time users to sophisticated repeat visitors. Another issue I will bring up is whether we should be building systems that are as human-like as possible, or whether we should be considering alternative design strategies. I will also briefly discuss plans underway through the W3C, the VoiceXML alliance, and proprietary solutions already available that deliver a WWW-like interface through a speech-only interface.
Throughout, I will support the discussion with case studies, try to show you how we actually do things at SpeechWorks, and I will provide demos of some of our deployed systems. If you'd like to try one of systems on your own time before or after the talk, I would suggest:
- United Airline Flight Info: 1.800.241.6522
a. choose option 1 from initial menu to get our
spoken flight info; it's all United offers now
b. try to get info on a flight; you might try to
get one arriving into San Francisco from New York's
John F Kennedy (JFK) airport at around 3 PM. Try
to see what happens if you mumble, or say the wrong
time or airport.
Several of our Voice Portal Applications (Bell South in Atlanta, Audiopoint in D.C., Foodline in New York, and stock quotes in Singapore) are described, with their live phone numbers, at:
Bob Carpenter is currently employed as a member of the dialogue team at SpeechWorks International, the leading provider of conversational speech recognition solutions for e-business. His two previous jobs were as a member of technical staff in the Multimedia Communications Lab at Lucent's Bell Labs and as an associate professor of computational linguistics at Carnegie Mellon University. He did his Ph.D. in Cognitive Science at the University of Edinburgh, but wrote his thesis during a year in the sunny environs of Stanford's Center for the Study of Language and Information. He enjoys logic, statistics, cognitive psychology, compiler design, high bandwidth, widescreen monitors, and lots of RAM, all of which he applies to his passion: natural language semantics.
Titles and abstracts for all years are available by year and by speaker.
For more information about HCI at Stanford see