Scientists have invented a “non-listening” smart speaker

Scientists have invented a "non-listening" smart speaker

PrivacyMic can filter audio information directly on the device, providing a higher level of security than encryption.

Today microphones are one of the most widely used electronic sensors in the world. According to some estimates, there are 320 million smart speaker microphones in the world, awaiting user commands. The problem is that they hear everything around them, including the conversations of the owners, and this can pose a privacy risk.

Experts from the University of Michigan proposed a solution to this problem – a development called PrivacyMic. Based on the Raspberry Pi, the device captures ultrasound at frequencies beyond the reach of human hearing. Working dishwashers, computer screens and even finger clicks generate ultrasounds at a frequency of 20 kHz or higher, scientists say.

The technology they developed analyzes the “ultrasound data” around and can identify home or office activity with 95% accuracy.

“There are many situations where we want home automation systems or smart speakers to understand what’s going on in our home, but that doesn’t mean we want them to listen to our conversations. We have found that you can get a system that understands what is happening, but never records audio information, ”explained University of Michigan professor Alanson Sample.

PrivacyMic can filter audio information (speech or any audio content) directly on the device, which provides a higher level of security than encryption or other measures aimed at protecting or restricting access to recorded audio data.

As part of the study, a team of scientists recorded the sound made when brushing teeth, vacuuming, computer and other activities, and then compressed “ultrasonic signatures” into small files, including key bits of information, cleared of noise in the range of human hearing. The team then ran an experiment where the participants listened to the audio collected by the device. As it turned out, none of them could distinguish human speech.

Currently, PrivacyMic exists only in the form of a prototype, however, according to scientists, to implement such technology in devices like “smart” speakers, only minor modifications will be required, in particular, software and a microphone that capture ultrasound.

Catch up on more stories here

Follow us on Facebook here

Leave a Reply