Engineers developed a new solution that could improve the protection of homes and individuals from unwanted data collection carried out continuously by smart devices. The study's main objective is to assist people in evading artificial intelligence systems such as Alexa, Google Assistant, or Siri from blatantly spying on their users.
According to the authors, the algorithm is a key solution to protect the privacy of people and their voices from their own devices.
Why Unsolicited Ads Show Up Creepily
Personalized ads popping out from screens on many devices frequently creep out many users. On many occasions, there were reports of individuals being shown an advertisement directly connected to the topic they have been talking about recently.
This is because of the high-quality microphones in almost every smart device scattered across the place. Many do not know that these pieces also serve as feed for the service providers and applications to listen to their customer's interests.
Moreover, microphones are commonly used by many neural networks and AI-based applications to analyze and learn the users' talk about giving more accurate suggestions.
Unsolicited ads are becoming a problem for the greater population that uses smart devices. But a new study, led by scholars from Columbia Engineering, presented a potential solution to cut the capability of these invasive systems and shield the users from uncontrolled data collection.
The research findings will be presented at the upcoming Tenth International Conference on Learning Representations (ICLR 2022).
ALSO READ : Explainable AI Accurately Labels X-Ray Images of Five Chest Pathologies Equivalent to Seven Human Experts
Blocking Smart Devices from Hearing, Spying on You
The new technique produces sound bites that can come in decibels as quiet as how people whisper. The innovative algorithm can be utilized by people anytime by just playing the random sounds, which would serve as obstructions that will screen normal words and conversations away from ad-driven feeds on phones and computers.
Columbia University's computer science specialist and co-author of the study Carl Vondrick explained in a press release that the most challenging part that their team encountered was making the system work as fast as possible.
The engineers' algorithm can block 'rogue' microphones from correctly hearing their users up to 80 percent of the time. Vondrick claims that the technique they assembled is the most accurate and fastest architecture they ever created.
Another stunning capacity that separates this innovation from other microphone blockers is that it can work wherever the rogue feed is. The privacy system will work well regardless of the spy microphone's location and the software it runs with.
The anti-microphone algorithm will camouflage an individual's voice as it travels through the nearby vicinity. This could make and protect the person's true voice from the listening devices without the need of concealing a casual conversation or talking too loud.
The pre-print of the study is available on arXiv, titled "Real-Time Neural Voice Camouflage."
Read also: Hypersonic Flight Race: NASA, ANL Develops Aircraft Engine Simulation Using AI and Machine Learning
Check out more news and information on Artificial Intelligence in Science Times.
