By: Tiffany Zhang

Many devices listen in on you when you don’t even know. That is why Columbia University researcher Mia Chiquier and her colleagues decided to make an algorithm to confuse the devices. On April 25, Chiquier shared the system her and her colleagues made at the virtual International Conference for Learning Representations.

To make the system outsmart smart devices researchers first needed to understand how such devices eavesdrop. Voice-activated smart devices use ASR to translate sound waves into text, Chiquier explains. The researchers created a system that adds sound waves to make it harder for the smart devices to understand the words you say. It “completely confuses this transcribing system,” says Chiquier.

For the system to work, it needs to predict what the person that is going to speak will say a few seconds before they say it. It then broadcasts chosen noises that would confuse the device’s interpretation of the words.

“Our system listens to the last 2 seconds of your speech. Based on that speech, it anticipates the sound you might make in the future.” Chiquier explained. The algorithm kept smart devices from hearing correctly 80 percent of the time.

Bhiksha Raj like their work but adds that the smart devices can still use the person’s voice to figure out things like their gender, health, mental state, accents, and physical size. With that information, the companies that own the smart device technology could exploit those features by showing users targeted advertising.