Partnering with CCNY on Project WHISPER

Partnership with CCNY

What if you wanted to hear what is being said inside a building two hundred meters away, and to hear that one conversation rather than everything inside the field around you, transcribe it, and save it? Those constraints together eliminate most of what people picture when they imagine long range listening. A parabolic dish at that distance collects a neighborhood.

The alternative starts from a property of sound that is easy to overlook. Sound is moving air pressure, and when that pressure reaches a pane of glass the glass moves. Not much. A window carrying ordinary conversation displaces by a small fraction of a micron, well below anything the eye or the hand registers. But it moves, and it moves in step with the voice that caused it.

Light can measure that. Aim an infrared laser at the pane and some of the beam comes back. Because the pane is creeping toward you and away from you as it vibrates, the round trip distance keeps changing, so the returning light arrives slightly out of step with a reference beam that never left the instrument. Combine the two and the mismatch becomes a voltage. Send that voltage to a speaker and you hear the room. The window has become a microphone that nobody installed.

Instruments that do this are called laser Doppler vibrometers, and they are ordinary commercial equipment used to check turbine blades and bridge decks and car panels for how they shake.

The selectivity problem takes care of itself. A vibrometer hears whatever surface the beam lands on, and at two hundred meters that spot is roughly the size of a person. Everything else in the scene is inaudible for the simple reason that it is not in the path.

Now the part that decides whether any of it is practical. Glass and painted sheet metal behave well, returning a clean orderly beam. Rough stucco or fabric scatters the light into a grainy pattern that shifts as the beam wanders, and that shifting shows up in the audio as hiss and dropouts. Working at a wavelength of 1550 nanometers helps, since that infrared comes back better from dark rough material and also cannot be focused onto the retina, which sets the eye safety budget generously. The improvement on diffuse surfaces runs to about twenty decibels. It does not make the problem go away. The published record shows clear speech recovered under a hundred meters on untreated surfaces, reaching three hundred meters only when somebody had put reflective tape on the target.

That gap defines the program. WHISPER exists to find which surfaces at two hundred meters still yield usable speech, and how much a modern signal processing stage can rescue from the ones that almost do.

Two hundred meters of air complicates things as well. Warm and cool pockets bend the beam continuously, so the spot wanders and the brightness flickers. Holding a millimeter class spot on one location at that range calls for arcsecond stability, which means a heavy tripod and motorized fine pointing, because any tremor in the mount comes back as sound apparently coming from the target. One aim for Project WHISPER is portability.