The walls around you may not need to be thin for a conversation to slip through. New research shows that everyday electronics could be leaking private audio, not via hacked software, but through physical signals that travel invisibly and quietly from your devices.
What drives this risk is the digital microphone, a tiny component found in nearly every modern device from phones to laptops. These components, which now dominate the electronics market, don’t just capture sound, they also emit subtle electromagnetic patterns. Under the right conditions, those signals can carry enough information for someone nearby to piece together what was said — even from behind a concrete wall.
A team at the University of Florida has demonstrated that speech can be reconstructed from these unintended emissions with surprising accuracy. Using radio receivers positioned up to two meters away, the researchers decoded speech that passed through thick structural barriers, including 25-centimeter concrete slabs. In some tests, their system transcribed digits with less than 7% error.
Each of these pulses generates a small burst of electromagnetic energy. When they fire rapidly, as they do during speech, they create complex signal harmonics that ripple outward. Microphone cables and circuitry, not designed to shield against such leakage, effectively become miniature antennas. What’s leaked isn’t noise — it’s structure-rich information that can be harvested with inexpensive gear.
By tuning into these frequencies and demodulating the signal, much like FM radio, eavesdroppers can retrieve not just raw sounds but intelligible speech. The emitted patterns preserve enough of the original waveform to distinguish voices, syllables, and even individual digits.
These weren’t controlled lab tests only. In one scenario, researchers placed a receiving antenna behind a standard office wall. Even without direct sight lines, the system extracted spoken numbers with over 95% accuracy.
Alarmingly, the microphones didn't need to be in active use for the attack to succeed. Tests revealed that apps like Spotify and YouTube can trigger microphone activity simply by playing content, even when mics appear disabled or muted. In several cases, microphones stayed live in the background, silently emitting data-rich electromagnetic patterns.
This accessibility changes the threat landscape. Corporate spies, stalkers, or bad actors could carry out surveillance without hacking, exploiting only the physical properties of common hardware. No special software is needed. No system breach is required.
Attempts to block the signal haven’t worked well. Signal filtering and noise suppression, typically used to protect audio fidelity, sometimes made these emissions easier to read. One approach, spread-spectrum clocking, did reduce the leakage. With just a 1% random variation in clock timing, signal readability fell dramatically, but applying this fix at scale would require redesigning the internal electronics of many devices.
For now, the researchers have shared their findings with affected manufacturers. While some companies responded with interest, at least one dismissed the concern, claiming compliance with regulatory standards and industry norms. Whether that response reflects technical confidence or a lack of urgency remains unclear.
The study adds a new dimension to the modern privacy debate. Digital surveillance already dominates the narrative, data mining, facial recognition, location tracking. But now, even the electronics sitting silently on your desk might be speaking volumes — whether you asked them to or not.
Read next: Wikipedia Halts AI Summary Trial After Editors Raise Concerns
What drives this risk is the digital microphone, a tiny component found in nearly every modern device from phones to laptops. These components, which now dominate the electronics market, don’t just capture sound, they also emit subtle electromagnetic patterns. Under the right conditions, those signals can carry enough information for someone nearby to piece together what was said — even from behind a concrete wall.
A team at the University of Florida has demonstrated that speech can be reconstructed from these unintended emissions with surprising accuracy. Using radio receivers positioned up to two meters away, the researchers decoded speech that passed through thick structural barriers, including 25-centimeter concrete slabs. In some tests, their system transcribed digits with less than 7% error.
How Digital Mics Became Unwitting Broadcasters
Microphones today often rely on pulse-density modulation (PDM), a digital encoding system that replaces analog waveforms with high-speed sequences of binary pulses. The more intense the sound, the denser the sequence of pulses. It’s efficient and compact, but it has a side effect.Each of these pulses generates a small burst of electromagnetic energy. When they fire rapidly, as they do during speech, they create complex signal harmonics that ripple outward. Microphone cables and circuitry, not designed to shield against such leakage, effectively become miniature antennas. What’s leaked isn’t noise — it’s structure-rich information that can be harvested with inexpensive gear.
By tuning into these frequencies and demodulating the signal, much like FM radio, eavesdroppers can retrieve not just raw sounds but intelligible speech. The emitted patterns preserve enough of the original waveform to distinguish voices, syllables, and even individual digits.
Vulnerabilities Found Across Devices
The team tested microphones from five popular manufacturers, including Knowles and STMicroelectronics, and found all to be susceptible. Recognition systems built on machine learning were able to match spoken digits to text with over 98% accuracy. Voice identification across speakers also remained consistently above 97%.These weren’t controlled lab tests only. In one scenario, researchers placed a receiving antenna behind a standard office wall. Even without direct sight lines, the system extracted spoken numbers with over 95% accuracy.
Alarmingly, the microphones didn't need to be in active use for the attack to succeed. Tests revealed that apps like Spotify and YouTube can trigger microphone activity simply by playing content, even when mics appear disabled or muted. In several cases, microphones stayed live in the background, silently emitting data-rich electromagnetic patterns.
Surveillance Made Cheap and Easy
Unlike high-end cyberattacks that require deep technical knowledge and costly equipment, this technique runs on pocket change. With basic copper tape, an amplifier, and a radio receiver, the researchers recreated speech capture systems that worked almost as well as professional gear. In several trials, even the stripped-down setup yielded speech recognition scores over 94%.This accessibility changes the threat landscape. Corporate spies, stalkers, or bad actors could carry out surveillance without hacking, exploiting only the physical properties of common hardware. No special software is needed. No system breach is required.
Attempts to block the signal haven’t worked well. Signal filtering and noise suppression, typically used to protect audio fidelity, sometimes made these emissions easier to read. One approach, spread-spectrum clocking, did reduce the leakage. With just a 1% random variation in clock timing, signal readability fell dramatically, but applying this fix at scale would require redesigning the internal electronics of many devices.
A Physical Threat to Digital Privacy
The implications stretch far beyond personal security. If office laptops or smart home devices can radiate speech data through walls, then sensitive meetings or private calls could be exposed to outsiders — without malware, wiretaps, or physical intrusion. The attack exploits the underlying behavior of the hardware itself.For now, the researchers have shared their findings with affected manufacturers. While some companies responded with interest, at least one dismissed the concern, claiming compliance with regulatory standards and industry norms. Whether that response reflects technical confidence or a lack of urgency remains unclear.
The study adds a new dimension to the modern privacy debate. Digital surveillance already dominates the narrative, data mining, facial recognition, location tracking. But now, even the electronics sitting silently on your desk might be speaking volumes — whether you asked them to or not.
Read next: Wikipedia Halts AI Summary Trial After Editors Raise Concerns
