Alexa heard what you did last summer – and she knows what that was, too: AI recognizes activities from sound Boffins have devised a way to make eavesdropping smartwatches, computers, mobile devices, and speakers with endearing names like Alexa better aware of what's going on around them. In a paper to be presented today at the ACM Symposium on User Interface Software and Technology (UIST) in Berlin, Germany, computer scientists …
But it could be fun trying to confuse Alexa.
Many moons ago a friend got a BBC sound effects tape and had fun with the message on his ansaphone - sounds of police sirens, helicopters and shooting in the background, to a gentle voice-over of "I'm afraid I can't get to the phone just at the moment, please leave a message"
Would Alexa call the plods if it heard shooting?
Forgive my ignorance of the state of the art - But can it distinguish "live" sounds from a TV or radio?
(I read the article twice, and all the comments, and no-one seems to have addressed this elephant in the room)
On the other hand, this could be incredibly useful for relaying the wife's softly-spoken orders - usually issued when I'm in another room/beside a boiling kettle or running tap/wearing headphones.
"Forgive my ignorance of the state of the art - But can it distinguish "live" sounds from a TV or radio?"
The article did say they trained it with sound effects from Hollywood, so my guess is no. Or alternatively yes, except the sounds from the TV will be taken as being "real", but the "live" sounds are obviously fake.
Just because you can do a thing it doesn't mean you should."
And for that matter, who do something you don't even need to do?
"The researchers suggest their system could be used, for example, to send a notification when a laundry load finished."
Surely your "smart" washing machine will email you when it's finished.
"Quite, and the researchers saying 'this should all be doable locally as people don't really want to be spied on' is like saying 'Lions don't really want to hurt people' - but they have to eat, right?"
Sure, but put away your paranoia for a instant and remember this is CMU saying this, as in the people that developed CMU Sphinx, and pocket sphinx. Which today is about the best bet for a local only speech recognition system as your able to compile/build at home. For that they should be at least given a small amount of the benefit of the doubt that they were being sincere.
I should know, I've been making my own smart speaker that is local network only to control some local network only home automation devices and currently I have pocketsphinx running on a beaglebone black but have some tuning issues to overcome...
Of course amazon et all will pish all over their good priniciples and aims, but there's still a small chance to carve out out a niche for privacy respecting alternatives because of these guys/gals work.
"There is no way to recover the audio"
...until another team of researchers discovers a way to contextually reconstruct spoken phonemes with 99% accuracy.
Great "use cases" guys. Alerting you when your washing machine is done? There's already a 100dB buzzer on it for a reason. And, correct me if I'm wrong, but they already make wi-fi enabled washing machines if you want it to post to your Twitter feed.
I'm sure these crazy inventors would be completely mystified by the feeling of revulsion that I get at the very thought of having such a device in my home.They are living in a completely different world.
THIS^ We keep hearing about more and more ways that these "boffins" are trying to come up with better ways to spy on people. Why don't all you twats of this sort out there work on something useful or, failing that, just fuck off and die?
Any device in my home that turns out to have a microphone on when I'm not specifically talking to it will get fixed with a hammer.
Currently, as a retired boffin, I feel that if we were in a position to hold the world to ransom with nuclear weapons in a volcano, it would be an improvement. We'd simply demand that all the politicians stop fighting and promoting stupid policies, spend the defence budgets on space exploration, fixing climate change and food security, and get on with doing something more interesting.
We keep hearing about more and more ways that these "boffins" are trying to come up with better ways to spy on people.
That research will happen regardless. It's much better when it's made public, as in this case, than when some giant corporation keeps it to itself. This way we at least know what's been done in the field, and can consider how to use or counter it.
Sorry, they are living in the real world. It's here and growing. You are the one left behind.
I just saw a youtube ad for Smart cameras for your home, billed as a must have for security of your home....
Lots of enticing pics of well-furnished empty rooms. It's a damn potential showroom and shopping catalogue for would be home invaders.
Tap that, and you can case a nice house from the safety of your own, and know when and how long the residents will be out and how vulnerable they are....
We've long gone beyond real world, we're in a fantasy at this point
It's a good thing for them, because they will be able to sell it for good money to Google/Amazon who will employ it to improve their profiles.
(And obviously also for those who will use it to collect "leverage" on certain people.)
But, as the article notes, we're just a small bunch of old fogeys always complaining; The younger generation is quite adamant about their right to be spied upon. Oh well.
"The younger generation is quite adamant about their right to be spied upon"
they have grown up on reality TV... and post their mundane activities on social media like anybody else cares about their bowels or whatever.
yeah, and if one o' those hears ME sneeze it'll sound like... "Ah, Ah, AH SHIT!!!" [yes I really do that, I hate sneezing - it interrupts what I'm doing]. I wonder if it'd call an ambulance, thinking it means "I fell and can't get up" (and of course THAT introduces the topic of false alarms generated by a big-nanny spy-on-your-life system). And no, I don't need Alexa to order me some narcoleptic cold medicine either.
"Harrison said a related project called Vibrosight, which involves using a laser to measure physical vibrations of an object to determine what it's doing, has already achieved sufficient accuracy for deployment."
Vibrosight? ..really.. Does it need a couple of AA batteries?
A couple of hours work in the audio editor and quick and easy distribution on YouTube can convince Jeff's little snoopbot that most people live in bizarre alien infested factories with opera singers, jet aircraft, laser guns and church bells, where doors slamming and creaking reveal populations of sinister schoolchildren, marching bands and herds of geese.
"On Tuesday, August 24th, I logged Mr. and Mrs. Marketing Hack having marital relations. Based on the excited chipmunk sounds Mrs. Hack made during this encounter, I sent a recommendation to our ad servers that in the future she be served with the commensurate battery of online pet store advertising."
"On Tuesday, August 24th, I logged Mr. and Mrs. Marketing Hack having marital relations. Based on the excited chipmunk sounds Mrs. Hack made during this encounter, I sent a recommendation to our ad servers that in the future she be served with the commensurate battery of online pet store advertising."
Well, from what I've seen of these annoying "targeted ads" that just show you things you've already purchased or own already, I predict that Alexa will just show Mrs. Marketing Hack more of the same old same old, which in this case would be ads for Mr. Marketing Hack.
"What CMU's comp sci types have added is a sophisticated sound-labeling model trained on high-quality sound effects libraries, the sort used in Hollywood entertainment and electronic games."
I wonder if they realise that those sorts of sound libraries are often built with fake sounds? Look up what it is that foley artists actually do to to create sound effects. Some examples from Wikipedia -
* Corn starch in a leather pouch makes the sound of snow crunching.
* A heavy staple gun combined with other small metal sounds make good gun noises.
* Burning plastic garbage bags cut into strips makes a realistic sounding candle or soft non-crackling fire when the bag melts and drips to the ground.
That last one is fun. Is the device hearing you burning some candles for a romantic interlude, or is your kitchen garbage bin on fire? Having the fire brigade turning up at your tenth anniversary dinner might not be a good idea.
* Frozen romaine lettuce makes bone or head injury noises.
* Canned dog food can be used for alien pod embryo expulsions and monster vocalizations.
Soooo, is your home under alien attack, or are you feeding your dog?
I wonder if they realise that those sorts of sound libraries are often built with fake sounds?
Irrelevant. They build the model, then they test it against the gold standard, which in a case like this is typically human judges.1
If the system does well relative to the standard, then it doesn't matter whether it was trained on authentic or synthetic data. The proof of the pudding is in the eating.
1Hopefully a representative pool of them, tested using a methodologically-sound approach, so the baseline is useful. I'm just assuming that here.
I'm trying to find words to express how completely unacceptable I find these new features, and failing miserably. Who the f**k decides that these kind of features are 'acceptable' ? Who decides that listening and trying to make sense of people's private conversations to sling another ad (yeah right) is an acceptable activity for a business.
And what's even more bewildering is that the American public, with their endless rhetoric about their right to bear arms to protect their constitutional freedom accept this behavior from their companies.
Someone should maybe explain to them that the amendment providing them this freedom was not thought up so kids could walk into their schools and start murdering their fellow pupils because they wear the wrong trousers, but to protect them from bent governments and people trying to take away their right to self determination.
I would expect a small army of minigun wielding freedom fighters would've taken up position in front of Amazon, Google et al HQ's by now.
I'm pretty much convinced that the likes of Samsung, Huawei, Nokia and every other phone maker have this capability built into their handsets but are just not telling us.
Pete
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
