Abstract
Acoustic signal has been recently adopted for contact-free hand gesture recognition due to its fine-grained sensing granularity and wide availability of microphone and speaker in consumer-grade electronic devices such as smartphones. However, a very limited sensing range constrains acoustic sensing to application scenarios where users interact with devices in close proximity. In this paper, we improve the range of acoustic sensing and demonstrate the feasibility of enabling room-scale hand gesture recognition using commodity smart speakers. We develop a series of novel signal processing techniques and implement our system on two commodity smart speaker prototypes with different numbers of microphones. Extensive evaluations are performed in three different environments with 1440 gestures collected from 16 participants. Experiment results show that our system can significantly increase the sensing range from 1 m to 4-5 m. In the challenging scenario where the user is 4 m away from the smart speaker and there is strong interference, the achieved gesture recognition accuracy is still higher than 90%.