We seek to test the hypothesis that text-to-speech(TTS) navigation systems can adequately guide pedestrians to unknown destinations in an unfamiliar city. Such systems bypass screen-based, multi-modal techniques and simply speak route following instructions incrementally into the pedestrian's ear piece. Due to errors in GPS positioning, uncertainty of user heading, poor map quality and potential communication and processing latencies, this becomes a surprisingly challenging task.

In our study, subjects are led on an unknown tour on the grounds of Umeå University. We evaluated both a human wizard controller as well as a simple decision-tree based controller and compared them to an ideal subject that knows the route. Results give support to our hypothesis that TTS-based navigation systems can adequately guide pedestrians. That said, our experiences point toward immediate and future improvements to make such systems more effective and agreeable.

All the software and data behind this work will be open sourced to encourage confirmation, replication and, ultimately, improvement upon our results.