Analysis of driver workload when using speech interfaces

Abstract

Speech interfaces are useful for improving the accessibility of in-vehicle devices as they are considered to be less distracting than visual interfaces. However, there is still considerable cognitive demand when using a speech interface, which may increase the chance of a traffic accident. This study investigates driver workload induced by various tasks, including a speech interface used for music retrieval. Driving data were collected from 523 drivers operating an instrumented vehicle on urban streets and highways. The drivers were given each of the following secondary tasks while driving: a) repeating random strings of four letters they heard through an earphone, b) interacting with a spoken dialogue system to retrieve and play music, c) being directed to an unfamiliar place by a human navigator on a cell phone, and d) reading aloud words seen on signboards while driving. After driving, each driver subjectively rated the induced workload for each secondary task from 1 (lowest) to 5 (highest). Analysis of the results of their subjective evaluations showed that talking to a navigator on a cell phone resulted in the lowest average workload (about 1.6), which was almost the same as the driving without any secondary task. Repeating random strings of four letters and retrieving music using a speech interface produced the highest average workload (about 2.5).