No abstract

Recently, there has been a surge in the popularity of voice-first devices, such as Amazon Echo, Google Home, etc. While these devices make our life more convenient, they are vulnerable to new attacks, such as voice replay. We develop an end-to-end system to detect replay attacks without requiring a user to wear any wearable device. Our system, called REVOLT, has several distinct features: (i) it intelligently exploits the inherent differences between the spectral characteristics of the original and replayed voice signals, (ii) it exploits both acoustic and WiFi channels in tandem, (iii) it utilizes unique breathing rate extracted from WiFi signal while speaking to test the liveness of human voice. After extensive evaluation, our voice component yields Equal Error Rate (EER) of 0.88% and 10.32% in our dataset and ASV2017 dataset, respectively; and WiFi based breathing detection achieves Breaths Per Minute (BPM) error of 1.8 up to 3m distance. We further combine WiFi and voice based detection and show the overall system offers low false positive and false negative when evaluated against a range of attacks.

Constructing a map of indoor space has many important applications, such as indoor navigation, VR/AR, construction, safety, facility management, and network condition prediction. Existing indoor space mapping requires special hardware (e.g., indoor LiDAR equipment) and well-trained operators. In this paper, we develop a smartphone-based indoor space mapping system that lets a regular user quickly map an indoor space by simply walking around while holding a phone in his/her hand. Our system accurately measures the distance to nearby reflectors, estimates the user's trajectory, and pairs different reflectors the user encounters during the walk to automatically construct the contour. Using extensive evaluation, we show our contour construction is accurate: the median errors are 1.5 cm for a single wall and 6 cm for multiple walls (due to longer trajectory and the higher number of walls). We show that our system provides a median error of 30 cm and a 90-percentile error of 1 m, which is significantly better than the state-of-the-art smartphone acoustic mapping system BatMapper [64], whose corresponding errors are 60 cm and 2.5 m respectively, even after multiple walks. We further show that the constructed indoor contour can be used to predict wireless received signal strength (RSS).

No abstract

No abstract