Designing and Evaluating In-car Speech User Interfaces

BY ANDREW KUN

According to the U.S. Census Bureau, Americans spend more than 100 hours a year on the road commuting. Given the large amount of time that people spend behind the wheel, and the increasing availability of computational resources that can now operate in a vehicle, many companies have been introducing in-car devices such as GPS-based personal navigation devices.

While interacting with aftermarket devices in cars is relatively new for most of us, police officers operating cruisers have had to deal with a multitude of brought-in devices for a while. Police cruisers are routinely equipped with lights, sirens, radios, radars and even laptops, all of which are considered essential tools to be used while an officer is driving.

But, how should drivers, be they morning commuters or law enforcement professionals, interact with the growing number of in-car devices? With more and more states adopting legislation barring the use of mobile devices in cars, speech user interfaces that allow drivers to control in-car devices are becoming increasingly common. Our lab at the University of New Hampshire has deployed the Project54 system, which integrates electronic devices in police cruisers and provides a speech user interface to them. The Project54 system is currently deployed in close to 1,000 cruisers in over 170 law enforcement agencies in New Hampshire as well as in over 150 cruisers in other states. Thus, officers of the New Hampshire State Police can issue voice commands to change channels on their VHF radios, turn on patrol lights, or initiate remote database queries for information on driver licenses and vehicle registrations.

How well do people drive while using a speech interface? The case of changing police radio channels is instructive. In a recent driving simulator study we evaluated the effects on driving performance of two modes of interaction with the police radio: using voice commands and using hardware buttons. We evaluated driving performance by calculating the variances of lane position, steering wheel angle and velocity (for all three variables a lower variance indicates better driving performance). We found that participants performed significantly better on the driving task when using voice commands. The image above shows our driving simulator.

Of course, drivers still need their hands to operate in-car speech user interfaces, since all commercial systems, including the Project54 system, require a push-to-talk (PTT) button. We are currently exploring a novel way to provide PTT functionality to the driver, via a PTT glove (see photo to the left shows our PTT glove). The PTT glove has push-buttons under the thumb and the index finger. The driver can press either of these digits against the steering wheel to initiate a speech command. We expect that the PTT glove is easier to use than a fixed PTT button since it does not require the driver to locate a button on the steering wheel - the PTT button is always under the driver’s thumb (or index finger). Preliminary driving simulator results are encouraging.

For driving performance to be satisfactory, drivers need to keep their eyes on the road. An in-car device, or the paper directions you printed out before your trip, can distract you from the road ahead with potentially dire consequences. In our latest study we compared the effects of three types of navigational aids on visual attention: paper directions, GPS-based navigation aids with voice prompts and LCD screens, and GPS-based navigation aids with voice prompts only. We evaluated participants’ gaze directions using an eye tracker installed in our simulator (a screen shot of the eye tracker output is shown below). Our results indicate that navigation aids that only provide voice prompts will reduce visual attention the least. However, taking away the LCD screen may be a tough sell: our participants preferred having the LCD screen over receiving driving instructions provided only via voice prompts. Finding out why this is the case is the subject of future work.

For reports on our progress please visit www.project54.unh.edu and www.eceblogger.com.