A Model to Predict Driver Task Performance When Interacting with In-Vehicle Speech Interfaces for Destination Entry and Music Selection.

Abstract

Motor vehicle crashes were estimated to be the eleventh leading cause of death in United States in 2009. Using a speech interface to operate infotainment systems while driving can potentially reduce driver distraction. Unfortunately, evaluations of driver interfaces are often too late to make changes. An alternative approach is to model driver task performance when using speech interfaces and to use the model to predict system performance early in design when changes are easier to make. The purposes of this research are to understand how drivers interact with speech interfaces and based on that knowledge, develop and validate a simulation model of how drivers interact with speech interfaces to aid speech-interface development. To develop the simulation model, a survey and a driving simulator experiment were conducted to identify how these tasks are carried out and the values for the process parameters. First, using a survey, frequency data for tasks and methods, and the content in user-generated databases were collected to assure that real tasks and constraints are considered in the simulation model. Next, a driving simulator experiment was conducted to understand how drivers perform destination entry and music selection and to determine the time drivers need to construct utterances, the errors drivers make, and the probability of correction strategies are used for each type of error. Half of these data were used to create the simulation model structure and provide the model parameters for entering destinations and selecting music using speech. Finally, the simulation model was validated for these two tasks using the second half of the data from the previous experiment. This research provides a model to predict user task performance with speech interfaces in motor vehicles. Use of this model supports the design of safer and easier to use speech interfaces in vehicles that can minimize eyes-off-road time and should reduce crash risk, and thereby protect public health. This model can be exercised to examine alternative speech interface configurations months before a physical interfaces is available for user testing when changes are easier to make, which saves time, reduces cost, and improves the quality of the interface produced.