A Sleeping Alexa Can Listen for More Than Just Its Name
spectrum.ieee.org
spectrum.ieee.org
Isn't that obvious though? Any internet connected device with a microphone has the capability to always listen to what you say. Whether they choose to or not is up to the manufacturer, and you're placing your trust in them. And it's funny to me that people freak out about smart speakers when your cellphone is within listening distance of you all the time.
the iphone has a low-power asic whose sole purpose is to listen for "hey siri" with relaxed confidence intervals. the real voice processing isn't done until the main cpu is woken by this asic. because of this, we can identify discrete modes of operation in the iphone, one of which is incapable of doing arbitrary "spying".
based on the article, it sounds like the alexa doesn't have this affordance. while unsurprising in this case, i don't think we must accept that anything with an internet-connected microphone has the capability to always listen to what you say.
I'm also not sure the article actually indicates what you're saying, it's kind-of vague. For example, it never states that anybody was able to actually program Alexa with a new wake-word, just the possibility of being able to create different signatures that could be used as wake-words. And the description of how they prevent things like the superbowl ad could all be done after Alexa detects the wake-word wakes up the main processor, but before Alexa actually responds to the user. I don't see anything in the article that indicates Alexa can't be using a separate chip dedicated to listening to the wake-word.
To that end, not to long ago a Reddit user claimed that Alexa did have two separate modes it operated in [0]. Which, it is entirely possible they could be lying, but you could always just open the thing up and take a look. But even if it does have a separate chip my previous point about the iPhone still applies, so besides power-usage I don't think it is extremely notable.
[0] https://np.reddit.com/r/Showerthoughts/comments/7m91u9/if_go...
Hell, if they were really crafty they could have it listen for a specific phrase or word, then send back as little as a single byte signifying that it was heard.
(I own an echo and a few Google homes, and I don't actually think this is happening, but the idea is kind of fun to chase)
What is the lowest bitrate codec and how does it compare to the plaintext version?
Voice codecs like AMR[1] (or GSM-HR) only need 2-3MB/hr. A trivial noise gate[2] that cuts out silence should allow a recording duty cycle under ~1%.
Amazon can easily store all the voice recording they want.
> and process ... in real time
Processing isn't that expensive once you have a trained classifier. However, realtime processing often isn't necessary.
[1] https://en.wikipedia.org/wiki/Adaptive_Multi-Rate_audio_code...
Never mind the wake word; a noise gate (which is a trivial computation locally in either hardware[1] or software) means we are not talking about [every second of every day]. I'm sure Amazon/whomever can do better than a simple noise gate, too.
Using GSM-HR which records voice at 5.6 kbit/s (a bit over 700 bytes/s).
5.6 * 60 * 60 / 1024 / 8 = 2.4609375 MB/hr
Assuming a gate that only records 1% of the time, this is 2.4609375 * 0.01 = 0.024609375 MB/hr
= 25.2 kB/hr
= 604.8 kB/day
604.8 * 365.25 / 1024 = 215.72578125 MB/year
Assuming 40M devices[2], storing all the voice said around these devices for a year is only ~8 PB (8229.28 TB). Renting the necessary storage from Amazon AWS, from a brief look at their current prices[3] costs at most $200,000/month ("standard storage").Yes, the storage costs add up over time, but the amount of data needed to record all the voice said around every Alexa in the world is trivially within Amazon's storage capability.
[1] voice activated tape recorders have been around for decades
[2] https://techcrunch.com/2018/01/12/39-million-americans-now-o...
Listening for words is power-intensive.
For me the most solid guarantee is the lack of a business case. Why risk the huge fallout of when this spying is inevitably uncovered when the profit to be had of listening in to people’s private conversations at scale is small. Individual private conversations might carry huge value, but discerning those from casual conversation at scale is next to impossible.
https://ryanplant.net/love.wav
Its only mistake was missing the "you" in "you want to be."
Codec2 remains fairly clear down to human ears down to 2.4 Kbit/sec but I had no luck getting it recognised, even with much simpler and clearer samples.
This individual anecdote tells you nothing about the state of the art, but I wanted to note it anyway because I was astonished at how well 6 Kbit/sec is handled.
But the love.wav gives a bit rate of 768kbps, 6 sec of audio at 571 KB storage size.
Another poster also did the math on 5.6 kbit/s (700 bytes/s).
> 5.6 * 60 * 60 / 1024 / 8 = 2.4609375 MB/hr
But should that 5.6 not be 768 to get love.wav voice quality? What am I missing like a moron?
Thanks for the example and anecdote, something to actually grasp in trying to understand the problem.
You'd think so, but no. I tried to explain this to some co-worker back when the Samsung TV issue came out. They didn't understand and literally made me out to be some conspiracy theorist. Their thinking is/was 'Why would X be listening to me if I didn't press the button or say the special word?' Same deal w/the cellphones. I've stopped trying to explain.
If it had that, I'd buy one.
There would also be a hardware switch in the same series. Turn it off, and no power can reach the device, no matter what the software decides. Manufacturers could easily make the circuit testable, so if you tell the machine to start using its camera, it could test the circuit and tell you that the switch appeared to be turned off and show you a little diagram to help you find the switch in case you accidentally switched it off and didn't even know it was there.
Most manufacturers who claimed to have this feature would do it correctly, because they would have to assume that somebody would disassemble a unit and damage their reputations if they faked it.
That's a pain electronically since the brightness of the LED would vary depending on the current drawn by the camera and/or microphone. In order to not blow the LED, the circuit may have to be set such that the LED wouldn't be bright enough to notice reliably when the camera is in a lower power (but still functioning) mode.
You could easily get what you actually want by putting an LED in parallel with the power supply to the camera and/or microphone though. Then the LED would maintain a fixed brightness but wouldn't be able to be switched off unless the power to the camera/microphone is cut.
Separately, it's possible to have unpowered microphones. I suspect laptop microphones are of the unpowered type (and amplified elsewhere). It's still quite possible to have a circuit which is hardwired to be incapable of passing the microphone signal without lighting an LED.
I'm just saying that your "in series" demand is unrealistic, and instead you should prescribe what you actually want rather than specifying an electronic circuit that can't necessarily achieve it.
My bluray player spends a lot of time fruitlessly searching for an internet connection when I put a bluray in it, and eventually it gives up and just plays the disk. I see no reason to connect it to the internet so it can spy on me.
This may become a thing for 4k Blurays (since all 4k Bluray players are long since compliant with the over-the-internet key updating mechanism), but unlikely because most manufacturers still refuse to update anything after warranty ends.
Ergo, the Bluray industry doesn't want to tie themselves to a model where people have to buy new players every 3 years: people will just stop buying discs instead.
The other OTHER side of this is: the disc is dead. Most people own very few discs, only of their absolute favorite movies they want in a high quality copy or for the few people who are bothering with 4k content (since most people still do not have enough internet to stream it); so why rock the boat, just let the format play itself out.
But you're right that I'll only buy a disc for a special movie.
https://iot.stackexchange.com/questions/2382/is-the-amazon-e...
<tinfoil hat on>But this is only as long as you actually can verify or trust that the switch is actually a switch and not a device that pretends to be one. A transparent casing where one can visibly confirm the actuator operation is a good idea.</tinfoil hat off>
It's mostly just a fiendishly annoying effect in high-sensitivity test equipment.
If Amazon advertised it as a physical switch, and it was not, that is fraud and actionable.
A physical switch is still a technically a physical switch, even if it's not connected to anything. It would come down to how they are describing what it does, and what guarantees they make (or not.) Do they have any, or is this discussion entirely hypothetical?
The switch doesn't need to be part of the main board --- it could be located next to the mic, so it's trivial for anyone with even the most basic electronics knowledge to see that it does what it says it does.
http://www.instructables.com/id/How-to-Make-an-Amazon-Echo-W...
On a Raspberry Pi or similar machine, you can hook the logic up to anything; a button, a switch, an infrared remote, whatever.
Axe should do it.
If we do dead-listing style analysis with teardowns and eeprom dumps, it can only tell us what it was listening to at that time.. and sometimes not even that.
How do we know what [.0076162328 3.27617819 91817.111121123 -65.2129] is listening to, especially when combined with complicated neural network weights? We can only check and verify every possible input, and that's effectively infeasible.
We chose not to buy these types of devices for our house, because of the great deal of unknowns.
But the same can be said of the tracker devices residing in many pockets. Some call it a cell phone or mobile phone, but its primary job and the job it does most of the time is to geolocate or track itself. And a tracker's mic has no indicator when it's listening, or a way for the user to vet the captured data, or control where the captured data goes. Those trackers that also have cameras are even worse in this regard. And they all run non-free, user-subjugating, proprietary software. So none of them respect a user's software freedom or privacy. I suspect that those with user-unmodifiable batteries aren't truly ever off except for when all of the batteries run down. But most users never turn their trackers off anyhow, so that's a bit of a moot point.
What other freedoms can proprietors convince you to do away with in exchange for some minor convenience or possibly a little innumerate fear-baiting (such as "you'll want this tracker in the case of an emergency!").
-----------------
I can figure out where people are talking spatially.
I can figure out how people are moving around when speaking.
I can voiceprint people.
I can cancel voices individually, or mute all noise past a certain distance from the center of the mic array.
I can 3d map rooms by recording over time how people talk and how they move.
I can connect multiple nodes together to create wider areas of coverage of above.
I can convert all sounds to text if appropriate. I can send tiny messages that include voiceprint, position, and content.
I can tell how many people are in your house at any one time.
I can tell if you have friends over, and when. If they have similar system, can collaborate backend data across all of above
------------
So yeah, there's plenty of things people aren't thinking of. I doubt they even have the experience to know how possible it is with open source tooling and cheap hardware. But all of these things are possible, and my guess is being done.
The distinction with phones is hard. Apple has stayed within the "we support privacy" aspect. And with things like limeSDR being cheap and common, wouldn't be as hard to attack and prove. Android is tricky, but with some of the anti-crapware systems and basebuild efforts going on, can be considered possibly trustworthy (if edited).
And in general, phones move a lot. Amazon tunacan/tomatocan doesn't. Just add a $1 accelerometer, and detection of movement is trivial. Phones can have this, but doing the above would be a magnitude harder.
https://www.cnn.com/2017/03/07/tech/amazon-echo-alexa-benton...
At the end of the day, you really can't trust anything completely, and the only real solution is to 1) trust Amazon (which doesn't seem like a great bet right now), or 2) don't own one.
Can you point us to facts please?
Because of TLS 1.2 and key pinning, we cannot see what's actually going on with all the data transfer between this hardware and the mothership/owner.
We can only do differential analysis, and say that X data was transmitted after doing X action. What that data entails, we do not know.
If you're happy with that, cool. I'm not.
https://www.theverge.com/2017/9/20/16340460/apple-ios-11-con...
I have an Android phone and I trust Google somewhat less in terms of personal-info-collecting than Amazon. :|
https://www.iot-tests.org/2017/06/careless-whisper-does-amaz...
With machine learning/AI systems these could potentially be more accurate than traditional home security systems.
I do know that my Nest Protects (smoke alarms) have motion detection though. Very handy nightlight in the middle of the night. Though if tied into the sole source of an alarm system I know it'd be a problem because my dogs can activate the nightlights in the Protects.
Pets really make home security hard, I'd bet.
I hate these things.
I've thought about making my own open source version using something like Jarvis.
Are there any systems out there that you can hook up via ethernet and hook up to your own server?
Things like having bedroom lights change color temp on time of day. perhaps stepper motor control of blinds. I really don't know why my lights need to be hooked up to the cloud via wifi for things like that.
They also have support for some wifi based things that are still local-only, but I tend to treat them as hostile and firewall them off, and even then I don't really like having them as they tend to be made by unknown manufacturers and have pretty shitty first party apps.
https://blog.mozilla.org/blog/2017/11/29/announcing-the-init...
Also, the solution to the smartphone problem isn't just to passively accept defeat. That will allow the problem to become worse.
https://news.utexas.edu/2017/06/26/the-mere-presence-of-your...
https://www.wired.com/story/our-minds-have-been-hijacked-by-...
"I'm sorry, I didn't get that. I'm sorry, I didn't catch that."
The "nothing to hide" arguments are irrelevant here -- people absolutely do have things to hide. This is where people seek advice about lawbreaking before they are ready or able to discuss with authorities. We can't have even the pretense that something could be listening in on these convos.
It's appropriate to blame the church for having the listening device.... but as technologists we know that off-the-shelf products don't give them much choice without technical proficiency/awareness to reconfigure the defaults.
Edit: and I think I interpreted your joke as you intended; no offense taken.
Imagine being in church, praying to God for the betterment of humanity, and then you hear Him answer you back, filling the church with His voice, but...it's just Cortana with an inane canned response.
It's funny, and not in an insulting way, except perhaps if you're on the team working on one of these products.
I also have a Google Home and it never, _ever_, answers by accident. "Hey Google" doesn't sound like other phonemes. Simple as that.
Also, I wouldn't have one if it wasn't given as a gift. I can't believe these are even for sale. They are clearly not ready for consumers. Nearly 90% of the commands I try to give result in Alexa not knowing or misunderstanding. Why should anyone pay to beta-test the product?
Do the Echo devices not have a hard mute switch?
I have a Google Home mini and it's usually on mute unless we want it for the same reasons, or to play trivia.
It is muted for weeks at a time, sometimes.
Of course I wonder how much of a hard mute it is. If you touch the thing while it's on and muted it will kindly remind you it's muted and can't hear you.
It would be nice to gain more insight.
I unplug it when I don't want to listen to music or put together a shopping list.
...which is when the device stores everything it hears, ready to send it to the mothership as soon as connectivity is reacquired.No idea whether it already does this but a small battery and a bit of flash go a long way when it comes to overhearing that which was not meant to be overheard...
I guess the pioneer of that was Microsoft and Windows 95.
- ad plays on tv
- alexa hears its name
- alexa sees that name contains the digital fingerprint
- (maybe alexa sends this to amzn servers so they know how may people r watching the sb & have alexa on)
- alexa stops listening for additional commands
therefore alexa is still only listening for its name
"
Echo is built on Texas Instruments DM3725 Digital Media Processor.
This TI SoC has two key pieces inside, first is ARM Cortex-A8 MPU, and the second one is TMS320DM64x+ DSP. The ARM core should be running Linux and the DSP is running firmware.
When idling, the ARM core is taken to lowest possible power state and Linux is completely suspended. At this time the DSP and 64KB On-Chip RAM are active. The DSP firmware processes noise coming in from the mics and attempts to identify if a keyword (e.g., Alexa) is spoken. As soon as it identifies there's a keyword, DSP sends an interrupt to wake up the ARM core which in turn resumes Linux.
"