Driving into the Future
Concept cars are visions of the future created by automotive manufacturers. Several 2003 concept cars also include visions of speech recognition. One of them, Ford Motor Company's Model U concept SUV is described as beginning Ford's second century of innovation. I interviewed Bryan Goodman of Ford Research and Advanced Engineering and Mike Phillips of SpeechWorks about the Model U's next-generation conversational system. Question: What was the purpose of building the Model U? Bryan: It was built as a technical and design exercise to show possibilities for several decades to come. It's shows forward thinking and the name, Model U, ties back to our roots. Mike: From our perspective the point of the project was to demonstrate that next generation interfaces are much more natural and that they can be used to control whatever you want them to. Question: Why did you include a speech-based conversational system? Bryan: The overall goal of the Model U was to create a positive view of the future. Part of that was personalization - using intelligence in the vehicle to enhance the driving experience. That includes enhancing both convenience and safety. Mike: The focus was less on the technology than on the user experience. Car makers want to put all sorts of functionality into the car and they need a way to do it that's cost effective, doesn't add too much to the dashboard and is safe to use. They think that speech plus some amount of display is probably the right way. Bryan: And, from a usability standpoint, we wanted a system that was easy to use and easy to learn - so that it could be useful whether it's a vehicle you've driven every day for years, a rental car or a brand new car you just drove off the lot. It also allowed us to push the envelope in terms of what user-interface technology is capable of in a fairly realistic system that's not light-years away. Question: We have voice interfaces in cars today. What makes the one in the Model U different? Mike: The current generation of speech systems work well, but they are command centered. We've added a lot more natural language so that you can say things in a much more flexible way. Bryan: The conversational interface we created allows control of a fairly large set of functions. You can get into a Jaguar today, push the button, and say "radio play." Rather than presenting you with a card that has 200 or more commands to memorize we wanted you to be able to learn to use the system in a matter of seconds. For example, if you want to change the seat temperature you don't have to know what's available or how to ask for it. You can say anything, like "I want to change the seat temperature." If that's all you say it'll take you through a dialogue to accomplish the complete task. It'll ask whether you want the seat to be warmer or cooler, and which seat - driver, passenger, both - you want to change. Once you've gone through that and know what's available you can start at the main menu, push the button, and say a single command, like "climate, please turn down the driver's seat temperature" or "please cool the driver's seat." You don't have to say anything specific. As long as you give it enough information about what you want it to do, it can do it. Question: How do you know you've covered all of the bases in terms of what a person might say? Bryan: You don't. There's a wide range of things that people can say. It was challenging and fun to think of those things and to do surveys about different ways people might refer to those functions. Question: What about handling noise? Bryan: You need what I call "speech done right," which is speech models that are created and tuned using real in-vehicle data. As far as I've been able to tell, whether the system is based on real in-vehicle data is what seems to differentiate systems that work well in real vehicles and those that don't. That's not difficult for automakers to do because we have proving grounds and test vehicles in the thousands. Question: Can we expect to see Model U conversational interfaces in Ford models soon? Bryan: No, its purpose was to express to people what we think Ford Motor Company will stand for in our next hundred years - and what could happen in a vehicle. None of it is production intent, today. Mike: This particular concept-car project wasn't necessarily tied to a vehicle program, but SpeechWorks is doing other things in the industry along the same lines that are tied to specific vehicle programs. I can't be more specific than that.
Dr. Judioth Markowitz is the associate editor of Speech Technology Magazine
and is a leading independent analyst in the speech technology and voice biometric fields. She can be reached at (773) 769-9243 or jmarkowitz@pobox.com.