Smart Speaker: Suspicious Event Detection with Reverse Mode Speakers

Abstract

Loudspeakers are transducers that convert electrical signals to sounds. However, it is well-known that in reverse mode, they can convert sounds to electrical signals. The paper studies the feasibility of designing an embedded system that detects suspicious events like gunshots or screaming by utilizing the reverse mode of loudspeakers. A proof-of-concept system was introduced, which can record the induced reverse mode signal in the inactive states of a speaker while providing the direct mode functionality in the active states. To examine the impact of the reverse mode’s distorting effects on the event detection accuracy, traditional audio event datasets were transformed into forms as they would have been recorded by speakers. Three different signal processing scenarios were introduced and evaluated by different methods and classifiers. The results suggested that in all three scenarios the reverse mode speakers could be used for event detection.