A power-indicator LED is wired directly to the supply rail. When the speaker draws current to play audio, the rail voltage sags — and the LED dims in time with the waveform. Nobody transmits anything. A photodiode across the room reads the brightness fluctuation and recovers the sound from the light. Your 1996 Mustang Christmas lights, weaponised from a parking garage.
the shared rail
The speaker and the LED share a power rail. The speaker draws current proportional to the audio signal. The rail voltage drops. The LED brightness tracks the voltage. The correlation is involuntary — no malware, no transmitter, no intent. The LED was only ever supposed to say "power on."
audio waveform→speaker current draw→rail voltage sag→LED brightness→photodiode→recovered audio
play → leak → recover
speaker
power LED
photodiode
press play — watch the LED breathe with the audio, then recover the sound from its glow alone
the signal is a short melody of pure tones. the LED brightness tracks the waveform because the speaker's current draw sags the shared power rail — the same physics as Christmas lights on an amplifier in a 1996 Mustang. the "photodiode" reads that brightness and reconstructs the audio.
what Ben-Nassi showed
The Glowworm attack (Ben-Gurion University, 2021) recovered intelligible speech from USB hub and speaker power LEDs at 15 metres (good intelligibility) and 35 metres (fair) using a Thorlabs electro-optical sensor and a telescope — through glass. No malware. No contact. The LED was wired to the rail, and the rail told on the audio. Tested on smart speakers, PC speakers, USB hubs, and Raspberry Pis. The fluctuations are invisible to the naked eye but strong enough for a photodiode.
what's authentic. the Glowworm attack (Ben Nassi, Pirutin, Gator, Zadov, Elovici, 2021) and its 15m/35m speech-recovery results are real and sourced. the power-rail physics — current draw modulates rail voltage, LED brightness tracks voltage — is exactly what happens. the demo computes a real audio waveform, applies a real voltage-sag model to generate the LED brightness signal, and recovers the audio by bandpass-filtering the brightness — the recovery is from the simulated optical signal, not a label lookup.
honest frame. the "LED" is a CSS div, not a photodiode reading physical light. the voltage-sag model is simplified (linear, no supply impedance or ESR modelling). real Glowworm needs a sensitive electro-optical sensor and often a telescope. the mechanism is exactly right — the ease and visual drama are overstated. real LED fluctuations are invisible to your eye; i exaggerated them so you can see the principle that your Christmas lights already showed you.
GLOWWORM · optical TEMPEST — sound recovery from a power LED