“...designed to appropriate any voice activated device.”
January 16, 2019 11:09 AM Subscribe
Project Alias By BjørnKarmann “Alias is a teachable “parasite” that is designed to give users more control over their smart assistants, both when it comes to customisation and privacy. Through a simple app the user can train Alias to react on a custom wake-word/sound, and once trained, Alias can take control over your home assistant by activating it for you. When you don't use it, Alias will make sure the assistant is paralysed and unable to listen by interrupting its microphones. Follow the build guide on Instructables or get the source code on GitHub.” [Vimeo]
The need for 3D printing and a soldering gun seem like a bit of a barrier to entry. FWIW I have a couple of Google Homes I use to listen to the radio (stations and channels that can't be picked up locally over FM) mostly. While I would definitely use this hack if it were a bit easier, on the other hand I already use Gmail, Chrome and an Android phone, so Google knows most things about me. Sigh.
posted by JamesBay at 11:48 AM on January 16, 2019 [1 favorite]
posted by JamesBay at 11:48 AM on January 16, 2019 [1 favorite]
I like that it looks like an alien brain slug.
posted by tobascodagama at 11:58 AM on January 16, 2019
posted by tobascodagama at 11:58 AM on January 16, 2019
It wouldn't surprise me if enterprising individuals began selling these on Etsy or something similar.
One thing that I just thought of, though: Project Alias relies on the ability to obfuscate the speaker's ability to discern voice by playing noise. The noise is more sophisticated than I thought it'd be--it's not just white noise, but a sort of vocal camouflage, various conversation-esque sounds pastiched together to form something that sounds like the background noise from a television, but entirely unintelligible.
The thing is, both Amazon and Google's voice recognition systems are cloud-based. In theory, given enough poisonous input and sufficiently advanced algorithms, they could be trained to filter out the noise and reliably recognize "real" input, even if that input isn't intended for your smart speaker's ears.
Is Project Alias actually just the opening salvo in some kind of voice recognition arms race?
posted by chrominance at 11:59 AM on January 16, 2019 [9 favorites]
One thing that I just thought of, though: Project Alias relies on the ability to obfuscate the speaker's ability to discern voice by playing noise. The noise is more sophisticated than I thought it'd be--it's not just white noise, but a sort of vocal camouflage, various conversation-esque sounds pastiched together to form something that sounds like the background noise from a television, but entirely unintelligible.
The thing is, both Amazon and Google's voice recognition systems are cloud-based. In theory, given enough poisonous input and sufficiently advanced algorithms, they could be trained to filter out the noise and reliably recognize "real" input, even if that input isn't intended for your smart speaker's ears.
Is Project Alias actually just the opening salvo in some kind of voice recognition arms race?
posted by chrominance at 11:59 AM on January 16, 2019 [9 favorites]
Chrominance - we've been there before! In WW2, Bell Labs and Alan Turing used two identical records of noise to encrypt audio. One end would transmit the voice along with the noise (carrier wave), and the other end would subtract the noise. https://en.wikipedia.org/wiki/Secure_voice
posted by rebent at 12:34 PM on January 16, 2019 [2 favorites]
posted by rebent at 12:34 PM on January 16, 2019 [2 favorites]
A Google Home will recognize multiple voices, this kills that ability.
posted by zeoslap at 12:36 PM on January 16, 2019
posted by zeoslap at 12:36 PM on January 16, 2019
The best bit, for me, is how it calls bullshit on google/amazon's insistence that there's any real reason you can't give your device a unique wake word.
posted by phooky at 12:37 PM on January 16, 2019 [2 favorites]
posted by phooky at 12:37 PM on January 16, 2019 [2 favorites]
Brilliant! A way to stop google/amazon collecting data from everything you say!
posted by Dr Ew at 12:41 PM on January 16, 2019
posted by Dr Ew at 12:41 PM on January 16, 2019
This is a product that is designed to mitigate the risk of bringing a potentially useful device that you do not trust into your home. That is a little horrifying and I want one.
posted by CaseyB at 1:06 PM on January 16, 2019 [5 favorites]
posted by CaseyB at 1:06 PM on January 16, 2019 [5 favorites]
is how it calls bullshit on google/amazon's insistence that there's any real reason you can't give your device a unique wake word
Yeah this random project with absolutely no data really shows that. Of course its possible to listen for random words, the problem is dealing with all the edge cases / false activations / etc. There's about 0 chance they solved that.
posted by thefoxgod at 1:27 PM on January 16, 2019
Yeah this random project with absolutely no data really shows that. Of course its possible to listen for random words, the problem is dealing with all the edge cases / false activations / etc. There's about 0 chance they solved that.
posted by thefoxgod at 1:27 PM on January 16, 2019
I would totally be into voice-activated smart home stuff if the voice processing didn't happen in the cloud.
posted by grumpybear69 at 1:46 PM on January 16, 2019 [2 favorites]
posted by grumpybear69 at 1:46 PM on January 16, 2019 [2 favorites]
In WW2, Bell Labs and Alan Turing used two identical records of noise to encrypt audio.
Right! So I feel like the first obvious thing for Google and Amazon to do would be to subtract the looped audio clip Project Alias uses for noise. Which then starts the arms race, as Alias figures out harder-to-crack methods of obfuscating voice, and Google/Amazon figure out ways around that protection.
First thought: I wonder if there's some way you could claim to a court that Alias's methods are a form of DRM that Google and Amazon are illegally trying to crack. Because that'd be hilarious.
Second thought: the second Google and Amazon decide to crack Alias in this way is the second Google and Amazon overtly declare war on the user's preference to be unmonitored unless asked, and that feels like a special line in the sand for some reason.
Third thought: Alias will probably never be popular enough for Google and Amazon to care, so I guess we're safe for now.
posted by chrominance at 2:14 PM on January 16, 2019 [2 favorites]
Right! So I feel like the first obvious thing for Google and Amazon to do would be to subtract the looped audio clip Project Alias uses for noise. Which then starts the arms race, as Alias figures out harder-to-crack methods of obfuscating voice, and Google/Amazon figure out ways around that protection.
First thought: I wonder if there's some way you could claim to a court that Alias's methods are a form of DRM that Google and Amazon are illegally trying to crack. Because that'd be hilarious.
Second thought: the second Google and Amazon decide to crack Alias in this way is the second Google and Amazon overtly declare war on the user's preference to be unmonitored unless asked, and that feels like a special line in the sand for some reason.
Third thought: Alias will probably never be popular enough for Google and Amazon to care, so I guess we're safe for now.
posted by chrominance at 2:14 PM on January 16, 2019 [2 favorites]
Why not have a user-definable or pseudo-random "seed" to modulate the obfuscating audio? Would make poisoned input harder to counter.
posted by porpoise at 4:39 PM on January 16, 2019 [1 favorite]
posted by porpoise at 4:39 PM on January 16, 2019 [1 favorite]
I would totally be into voice-activated smart home stuff if the voice processing didn't happen in the cloud.I too would be entirely "into" this, so long as the code was simple enough that I could understand it and I could verify that it was in the actual code that was running on the device.
And as long as we're wishing, let's add a unicorn that poops hot fudge sundaes.
posted by Gilgamesh's Chauffeur at 4:55 PM on January 16, 2019 [1 favorite]
If you're worried that Googazon will subtract the noise from this to reverse-engineer, it's relatively easy to fix. The GitHub repository has a single file called noise.wav, which is the 1 minute of random noise that is used to block out the listening. Replacing it with your own minute or more of speechlike noise seems fairly trivial relative to the tech involved in setting this up, at which point even if Googazon decides to develop a filter for Project Alias, it won't be able to filter out your device.
Although it would be nice to be able to incorporate some randomness -- have it loop a mix of two random noise files of different lengths (say 59 and 61 seconds - or better yet, 179 and 181), but offset the second by a tiny random amount, for instance.
The other thing is, at some point if there's enough noise coming in then it becomes difficult to remove the added noise to find the background, depending on how good the microphone is. Sort of like if you added noise to a picture that darkened it, at some point the pixels are black and once all the pixels are black, you can't lighten it.
posted by Homeboy Trouble at 5:05 PM on January 16, 2019
Although it would be nice to be able to incorporate some randomness -- have it loop a mix of two random noise files of different lengths (say 59 and 61 seconds - or better yet, 179 and 181), but offset the second by a tiny random amount, for instance.
The other thing is, at some point if there's enough noise coming in then it becomes difficult to remove the added noise to find the background, depending on how good the microphone is. Sort of like if you added noise to a picture that darkened it, at some point the pixels are black and once all the pixels are black, you can't lighten it.
posted by Homeboy Trouble at 5:05 PM on January 16, 2019
I would totally be into voice-activated smart home stuff if the voice processing didn't happen in the cloud.
I too would be entirely "into" this, so long as the code was simple enough that I could understand it and I could verify that it was in the actual code that was running on the device.
And as long as we're wishing, let's add a unicorn that poops hot fudge sundaes.
All you need to do is try it out while disconnected from the internet. Or, better yet, never connected at all. No mythical beasts required.
posted by grumpybear69 at 8:51 PM on January 16, 2019
I too would be entirely "into" this, so long as the code was simple enough that I could understand it and I could verify that it was in the actual code that was running on the device.
And as long as we're wishing, let's add a unicorn that poops hot fudge sundaes.
All you need to do is try it out while disconnected from the internet. Or, better yet, never connected at all. No mythical beasts required.
posted by grumpybear69 at 8:51 PM on January 16, 2019
It's weird to imagine being just security conscious enough to make this but not enough to take it to the point of just keeping the mic disconnected within the device until the activation phrase is mentioned.
Currently it's an easy-to-install, easy-to-uninstall mod: To install, plop it on top of your appliance; to uninstall, pick it up. To add an electronic cutout to the mic probably would require some intensive modification of multiple surface mount components, which is... harder. (The Echo microphones are the things with green circles around them in this photo.)
posted by jjwiseman at 10:57 AM on January 17, 2019
Currently it's an easy-to-install, easy-to-uninstall mod: To install, plop it on top of your appliance; to uninstall, pick it up. To add an electronic cutout to the mic probably would require some intensive modification of multiple surface mount components, which is... harder. (The Echo microphones are the things with green circles around them in this photo.)
posted by jjwiseman at 10:57 AM on January 17, 2019
It's weird to imagine being just security conscious enough to make this but not enough to take it to the point of just keeping the mic disconnected within the device until the activation phrase is mentioned.
I actually do wonder if there would be any issues with converting this into something that just intercepts the USB power for the speaker and only provides it when Alias hears its activation phrase. Maybe the Home/Alexa speaker startup takes too long for that to be feasible or something.
posted by chrominance at 1:58 PM on January 17, 2019
I actually do wonder if there would be any issues with converting this into something that just intercepts the USB power for the speaker and only provides it when Alias hears its activation phrase. Maybe the Home/Alexa speaker startup takes too long for that to be feasible or something.
posted by chrominance at 1:58 PM on January 17, 2019
« Older Over here we call that a "Kaiju Pizza" | "A flock of sheep brings humanity and freedom to... Newer »
This thread has been archived and is closed to new comments
posted by evilDoug at 11:36 AM on January 16, 2019