Lamphone: Real-Time Passive Sound Recovery from Light Bulb Vibrations | Hacker News Reader

Lamphone: Real-Time Passive Sound Recovery from Light Bulb Vibrations

nassiben.com

125 pointssohkamyung30 comments

The graphs they present show the largest response with 110db audio, which is about as loud as a car honking right in front of you[1], but even still, to eavesdrop on corner offices from tens of meters away with nothing but a telescope and a video recorder is a huge feat.

[1] https://www.chem.purdue.edu/chemsafety/Training/PPETrain/dbl...

That’s to be expected, but they could still recover audio at vastly lower DB levels.

Pointing a telescope at tens of meters away will get you noticed pretty quick. Now one can sell lightbulbs with noise modulation to defeat it.

Yeah, not sure why they patented the attack, and not the defense.

It seems much easier to enforce a patent against technologies that defeat the attack.

From the article, it looks like one can build the surveillance device from a few readily available parts, and then hack together the signal processing software. That’s not even in violation of the patent if you don’t sell / distribute it.

The attack obviously works only at night, when the light bulbs are on. Pretty sure you can easily disguise it.

HN has a real weakness for cool surveillance technology.

Sure. I for one have a huge weakness for this kind of tech, because it's an archetypal hack - a clever trick to pull something out of a system, whether the system likes it or not.

It's also interesting because it's deeply grounded in reality. It's no gluing crappy APIs together to solve a "problem" twenty layers deep in someone's convoluted business shenanigans. It's using reflected light to learn about mechanical vibrations at a distance, to show that the information is there, despite what you'd hear from people who don't comprehend how reality works. It's a nice reminder that everything that happens interacts with everything around it, and information about it spreads out at the speed of light. Most side channel attack are like this, which is why I find them interesting.

If you want to fight side-channel attacks, you can't put your head in the sand and pretend they don't exist (which seems to be most people's preferred reaction). You need to publicize them.

I wonder if that has something to do with how many cybersecurity policies are based on these weird edge cases and not on 99.999% of vulnerabilities.

The technology and the research has always been cool. That doesn’t mean people are happy with the application of it.

What you’ll also find is that the people who like the surveillance could barely give a crap about the technology that enables it.

The Venn diagram here isn’t one that completely overlaps.

I have the impression (not least based on the original spectrograms) that the sound sources were quite loud and would have been annoyingly loud in the room, so your whispered conversations are probably safe for a while yet. But it's very impressive work nevertheless. I look forward to trying it out.

So what's the gold standard of whispered convos these days? Last I'd been paying attention it was standing in the surf.

I believe what you do is you record an hour of normal conversation and then play back dozens of timeshifted copies simultaneously. Any system is going to have a very hard time filtering out noise with the same statistical properties as the signal - including you. You might have to stand quite close together...

If I'm remembering the right book, I believe something like this was described in Anathem by Neal Stephenson (not crucial to major plot).

Cool idea, it seems like version 2.0 of the old trick of turning up the radio.

That’s brilliant

ASL in a faraday cage :)

More realistically, a bathroom shower; the water plus the echoes from the tile/glass in typical bathrooms are better white noise than any digital solution. Two pots boiling at different heats also works and you potentially get the benefit of nutritious meal. Many kitchens also have noisy extractor fans conveniently located right above the stove!

Working in film, you want sound to be realistic but viewers have to be able to link it to something on the screen. Those loud hums from store/restaurant refrigerators used to be the bane of my life because while they were often the authentic natural sounds of the environment they're rarely visible unless you are specifically shooting a scene in a meat locker or next to an airport.

If I was shooting a remake of The Conversation for the digital age and my secretive chatters weren't using high tech wizardry, I'd give them two am radios tuned to bible or foreign language channels. 3 overlapping sound sources recorded at a single point (without the direction resolution our ears provide) are possible to disentangle, but it's a ton of work. Speech is harder to filter than music.

If you're in real fear of aggressive surveillance but must communicate, pick the noisiest room (usually the kitchen), sit on the floor or under a table, and hold your conversation inside a large cardboard box with some thick blankets over the top. I would sometimes use this when I needed to record in a noisy environment where it wasn't practical to run wires.

Thanks for the movie. The kitchen suggestion reminds me of the joke:

- Did you know the soviets invented the Internet?

- What? Are you high?

- It's true: we used to sit in the kitchen in our pajamas, talking about everything but never getting anything done. The only difference nowadays is that the illumination is cold white instead of warm orange.

A russian take on https://xkcd.com/386/ "Duty Calls": https://www.youtube.com/watch?v=kOeAXKUK6aM

Those "White noise generators" make whispered conversations hard to hear...

Except you can design White noise which is pseudorandom rather than truly random, and then you can subtract the noise back out again...

If you trust Sox to be random enough, there's

    play -n synth whitenoise

if not, perhaps

    cat /dev/urandom | aplay

would suffice.

Then you can buy a white noise diode.

Except IC’s are the same size as discrete white noise diodes these days.

If thats the case, then they really should have pointed that out

Nah, you can solve that by just using more expensive tech (ie higher adc bit depth and faster sampling). The cool thing is that it's doable!

It depends on what the noise sources are. If you're ADC limited, sure, but if you're limited by something else--detector noise, air turbulence affecting the signal, variation in LED output, noise in the room--a better LED won't help.

Except it will a long way down. The random noise would have to match the characteristics of human speech for there to not be information that's possible to gather in an uncoded system.

That's an interesting claim. It's true in theory--the information is still there--but what's actually going to do the separation of the noise from the signal?

One would have to characterise the system and apply pretty rad filtering.

Actually hmm, I don't know if machine learning or neural networks or genetic programming have been tried on this, but it sure does sound like something they could work at!

Whoops, should be "a better ADC won't help."

Reply on news.ycombinator.com