This post is also available in: heעברית (Hebrew)

Are you experiencing online ads following you on topics you were just discussing with your friends or family? Today, microphones can be found almost everywhere in electronics, such as smartphones, watches, and televisions, which are always listening to our conversations. Artificial intelligence and neural networks are used to obtain additional information about us from our voice data. But what can be done to avoid this?

Engineers at Columbia Engineering have developed a system that creates sounds similar to whispers, creating a impenetrable barrier against spying on smart devices in any room and any situation. Different hardware can also be used with the system such as computers, smartphones, allowing privacy to be protected.

Researchers had to create an algorithm that disassembles neural networks in real time in order to avoid possible microphone spying, produce results while speech is being made, and also support different languages. The team developed an algorithm that uses “predictive attacks” to neutralize the possibility of monitoring the ongoing conversations in real-time, according to engineering.columbia.edu.

They developed an algorithm that is fast and accurate enough to keep spying microphones from picking up words 80% of the time.  This technology hides a person’s voice over-the-air so it’s not detected by hearing systems during real-life conversations. Detecting the characteristics of what a person will say next allows it to generate the right whisper at the right time to make spying impossible, thus providing an additional layer of privacy protection for the speakers. The research is another another illustration of how artificial intelligence may be utilized to safeguard us against its misuse.