When you think about clandestine listening devices, you might imagine a tiny microphone hidden inside a lamp. But Ben Nassi, a PhD student at Ben-Gurion University of the Negev, found that the lamp itself can be used to listen to conversations remotely and in real time.
He and his team used minute vibrations on a hanging lightbulb to reconstruct audio from within a room. They dubbed their creation Lamphone.
The primary question Nassi and his team set out to answer was whether a hanging lightbulb can be used as a microphone—a challenge since “lightbulbs were not exactly designed to be used as microphones,” Nassi said at this year’s virtual Black Hat conference.
Vibing Out
Sound, Nassi reminded the socially distanced audience, is just air moving through space. When it hits an object, it causes that object to vibrate. “By analyzing how the object responds to sound, the vibrations, with the proper sensor, can be recovered.”
On a lightbulb, this vibration is miniscule. It took a lot of lab experimentation for Nassi’s team to understand how the bulbs were vibrating in response to sound. Interestingly, the team determined that sound could be reconstructed from any E27 lightbulb—be it fluorescent, incandescent, or LED, but the LED had a higher signal-to-noise ratio.
To measure that vibration at a distance, Nassi’s team mounted an electro-optical sensor to a telescope. This would capture the light from the bulb, and translate changes in the amount of light into voltage, which could be fed into a computer.
To prove their system worked, Nassi’s team set up their telescope and sensor on a pedestrian bridge 25 meters away from the target office. Inside, a 12-watt bulb would serve as the “microphone.” The team had to clean up the signal, applying filters and a custom equalizer. Despite their experiment sitting on top of a train station, the team was able to capture vibrations on the bulb and reconstruct sounds played in the office.
Nassi’s team played two songs inside the office: Coldplay’s “Clocks” and The Beatles “Let It Be.” Comments on musical taste aside, the system successfully recovered the songs. True, they sounded quite a bit muffled, like something played very loudly over a phone, but they were unmistakable. It was even good enough for both songs to be successfully identified by Shazam.
The team was also able to pick up a speech played in the room, using a clip for US President Donald Trump. “Bear in mind, Lamphone is capable of understanding President Trump, which is a very difficult task nowadays.”
Burning Brighter
While Lamphone is a success, Nassi said there’s room for improvement. A telescope with a larger diameter lense would capture more light, Nassi explained. This might allow for lower volume sounds to be picked up by Lamphone.
The team also employed only basic audio-processing techniques. More advanced techniques could improve the system even more. “Deep learning is an option,” said Nassi, although he conceded that this may require a lot of data for training a deep learning system.
At the end of his talk, Nassi looked to how quickly similar attacks had progressed. Gyrophone, which used the gyroscope in a mobile phone to capture sounds, took just six years to become a practical threat to privacy. Perhaps in 2026, Nassi mused, some other PhD student would present an even more effective version of Lamphone at Black Hat. Here’s hoping we’ll be able to see it in person by then.
