How do you say "Okay, Google" in dolphin?
September 8, 2017 5:38 AM   Subscribe

Voice-controlled assistants by Amazon, Apple and Google could be hijacked by ultrasonic audio commands, according to researchers in China and the US. Amazon and Google say they are reviewing the claims.
posted by clawsoon (16 comments total) 6 users marked this as a favorite
I read about this this morning. I love my Neal Stephenson future.
posted by Literaryhero at 5:41 AM on September 8, 2017 [2 favorites]

Two teams said the assistants responded to commands broadcast at high frequencies that can be heard by dolphins but are inaudible to humans.

"Siri, order me some tuna."
posted by ricochet biscuit at 5:57 AM on September 8, 2017 [6 favorites]

Me: [ inaudibly gasps ]

Google: Save your review?
posted by srboisvert at 6:30 AM on September 8, 2017 [2 favorites]

So looping it through Jones is actually going to be a valid hacking technique now?
posted by radwolf76 at 6:36 AM on September 8, 2017 [7 favorites]

Solution: crappier microphones and/or a hardware lowpass filter on the input prior to processing. There is no reason those mics should be registering stuff coming in at 12-15k.
posted by grumpybear69 at 6:44 AM on September 8, 2017

"Alexa, so long and thanks for all the fish."
posted by drezdn at 6:51 AM on September 8, 2017 [3 favorites]

Solution: crappier microphones and/or a hardware lowpass filter on the input prior to processing. There is no reason those mics should be registering stuff coming in at 12-15k.

Not that simple. The mics are registering audible frequencies when only ultrasonic frequencies were used, because of a nonlinear effect present in the microphones themselves.

Say you have a signal with two frequencies (A,B). That signal passes through a nonlinear medium- say, the internal bits of Alexa's microphone. That nonlinearity creates new sum and difference frequencies A + B and A - B. The hardware then lowpasses out everything over 20kHz, so the digital-to-analog converter outputs just A - B.

The tricky thing for this technique is that the nonlinearity produces two new frequencies for every pair of input frequencies (this helps electric guitar chords sound awesome). That may constrain the audible-band signals they can produce, and maybe the software can try to notice when it receives a signal within those constraints.
posted by Jpfed at 6:58 AM on September 8, 2017 [8 favorites]

you could still get away with a tricker hardware filter before it goes to the baked-in listener chip. tricky, but these are the folks who baked in a deep learning chip for a hundred million dollars to get better latency on the "OK Google" or "Hey, Alexa" recognition
posted by hleehowon at 7:09 AM on September 8, 2017

I am sitting in an amazon warehouse different from the one you are in now.
posted by mhoye at 7:31 AM on September 8, 2017 [8 favorites]

Pha loves Pa.
posted by doctornemo at 8:21 AM on September 8, 2017 [1 favorite]

The Pha loves Pa link led me to John C. Lilly's body of work. I read most of what he wrote when I was in my early twenties. I took a lot of it to heart and mind. Reading again through his bio, I suddenly remember his work is bedrock to my take on living. I think that Programming and Metaprogramming in the Human Bio Computer, well I have to read that again, very soon. It is always amazing to rediscover agreements both made, and forgotten. Again, buy a computer monitor without cameras or microphones. Keep your phone, under a piece of paper, don't take your phone in your home, where you would not take a total stranger. Do not leave your phone by your bed, keep it covered or in a closed case, unless you are using it. Keep your computer turned off and the router unpowered when not using your computer. What? Watch what your kids are doing and listen to the tone of the content they view from a discrete distance. When your children become enraged, gaming, then ask them to shut things down.

In the human nervous system, high stress changes us drastically. Computer simulations artificially create ultra high stress, that your kids and grand kids would not normally encounter in the sane life you have , hopefully, created for them as best as you are able. Outsiders are turning them in emotional disaster victims on a daily basis, or worse, winners, obdurate bullies killing with a screen. Killing for pleasure, not what I had in mind for mine.
posted by Oyéah at 10:09 AM on September 8, 2017 [2 favorites]

The Story of Solomon Epstein
posted by jeffburdges at 11:38 AM on September 8, 2017

Yet another reason we should to learn to speak dolphin.
posted by homunculus at 3:20 PM on September 8, 2017

Wired Magazine for that last link. John C. Lilly for the wired concept.

Dr. John Cunningham Lilly (January 6, 1915 – September 30, 2001) was an American ..... The result will be that Earth's consciousness will awaken as people become linked nodes in The Wired network.
posted by Oyéah at 3:43 PM on September 8, 2017

I knew there was more to Gavagais' efforts to decode dolphin speech....
posted by Rabarberofficer at 7:46 PM on September 8, 2017

I read this the other day, and the paper itself along with the demonstration video are absolutely compelling.

It's not just that they're exploiting non-linear responses of the microphones before the signal hits the low-pass filter, but they also synthesized the users voice to generate an identifiable "Hey Siri" command to activate the voice control, by putting together individual phonemes captured by recording unrelated speech from the user.
posted by TwoWordReview at 11:00 AM on September 11, 2017 [1 favorite]

« Older A Wild Melodica Appears   |   And the "winner" is... Newer »

This thread has been archived and is closed to new comments