Spee-Navi: An Indoor Navigation System via Speech Crowdsourcing

We present a speech crowdsourcing-based navigation system, Spee-Navi, which is able to guide users to the target in ambiguous indoor environments. The proposed approach leverages the microphones in the commercial off-the-shelf devices to extract the semantic feature for space partition via Automatic Speech Recognition (ASR). Moreover, the reversed Dead Reckoning from the target area to the user is implemented to provide a shortest path without comprehensive indoor localization or even the floor plan. Spee-Navi collects a rich set of sensor readings during the course of a guider's walk, analyzes the location semantics of purchase areas using ASR when the guider stops, and then packs them into a navigation trajectory at the entrance. The followers, who have been provided an appropriate path on the basis of semantics, track the navigation trajectory and get prompt visual instructions. We detect both course-grained and fine-grained semantics, and conduct extensive experiments. The evaluation results show that Spee-Navi can navigate a user to the target area within 9 steps error.