TV anchor says live on-air 'Alexa, order me a dollhouse' – guess what happens next A San Diego TV station sparked complaints this week – after an on-air report about a girl who ordered a dollhouse via her parents' Amazon Echo caused Echoes in viewers' homes to also attempt to order dollhouses. Telly station CW-6 said the blunder happened during a Thursday morning news package about a Texan six-year-old who …
Until someone creates a voice controlled system that responds to the wake phrase "Computer" and responds in the voice of Majel Barrett Im not buying anything.
When that day finally comes Ill be making my family wear RFID tags simply so I can shout "Computer, Locate Commander Wife" - I might even rename my house to "The ship" so that it can reply "Commander Wife is not onboard the ship"
"Until someone creates a voice controlled system that responds to the wake phrase "Computer" and responds in the voice of Majel Barrett Im not buying anything."
Not sure if there's a Majel Barret voice for Garmin SatNavs, but you can set your own wake up word. I did set it to "computer" for a while but the word "computer" comes up on the radio often enough to make that a problem. Especially since the speech recognition for commands is actually quite basic so words which sound even a little like the command can cause actions once triggered. eg I discovered that the vowel sounds "O" as in "comp", "OO" as in "ook" and "A" as in "sat" spoken in the right cadence to sound vaguely like "computah" would trigger voice mode. Once in voice mode, various commands may then be acted on by normal conversation/sounds/voice on the radio, but at least most require an eventual "Yes" to complete.
Since our hearing is generally peaking at 20khz, seems most sound repro systems don't go much above that.
But, assuming (no idea) live human voices actually carry recognizable higher harmonics, could devices not use special microphones that recognize their absence and therefore infer that it's listening to a recording or broadcast?
As an added bonus, folks with Monster cables or the $30K stereos could still get dinged!
But, assuming (no idea) live human voices actually carry recognizable higher harmonics, could devices not use special microphones that recognize their absence and therefore infer that it's listening to a recording or broadcast?
I have a vague memory of hearing of something like that being worked on. Whether it was somewhere like TV/Movie, SciFi book, or something non-fiction I'm afraid I couldn't say, but the idea was that certain frequencies weren't reproduced by electronic systems so if they weren't detected.
That said, I am sure that a sound source generating 1khz will produce harmonics and sub-harmonics the same whether human voicebox or speaker (no idea what freqs human voice ops at)
As an added bonus, folks with Monster cables or the $30K stereos could still get dinged!
Not necessarily true.. The speaker and the amp make the system, the cable is just, well, nothing but wire. I have friends who believe my speakers are fed by some expensive cabling throughout the house, just because it sounds right to their trained ears. Truth is.. I had a spare roll of mains cable... Oxygen-free copper with gold plating isn't going to carry electrons any differently, not over these distances (and aside from perhaps lowering the resistance of the cable, I doubt it would matter over any distance (potential interference aside), electrons is electrons)
But this a nice idea, and I cannot think of a more deserving bunch for some pranks based on "faithful reproduction" of certain sounds..
This post has been deleted by its author
Lots of you are saying voice will never be as good, but how can it get as good as text-based input unless we actually start using it? You can't improve something that doesn't exist or doesn't get used.
I personally love my Echo Dot and did the similar thing by ordering something unintentionally the first day I used it. Cancelled the order quickly and turned off ordering via voice (although there is a PIN function, so Amazon have thought about protecting it from being used by unauthorised users). It could be improved by alerting the Alexa app to prompt the purchase before accepting the order.
Again, it's illiterate tech users that don't do their homework and can't set up their products the way they need to secure it. All the docs are there and app is so easy to use.
The Amazon Echo/Dot products do. The option to turn off ordering is there in black and white (includes explaining the consequences). It's not like users have to find a "dip switch or jumper" to modify to get a different option setting.
Keep bashing the users why don't you.
I have a Echo Dot that sits on my living room coffee table. The other day I was watching CNBC business news show and they were doing a article on this subject. Every time they mentioned the wake up word: Alexa, the device would light up indicating that it too was listening to my TV. Of course, it would not understand any thing that followed but I thought it was amusing.
I still don't find voice interfaces much use beyond voice dialling in the car. I don't really see the point of them most of the time and you still look like a total moron barking commands into Siri or Google Assistant on a headset walking along the street or in public transport.
"Hey Siri .. play most embarrassing playlist ever!"
"OK Google: Where's clapham junction?"
"Hey Siri... do I look like a gimp talking to myself on a bus?"
"Yes! Yes, you do! I had been meaning to say that for years!"
If I'm in an area where I really don't want to take my smartphone out - I will use my watch to adjust play lists. Otherwise, I'll just use the phone (which most of the time as very few places are THAT threatening).
When voice activated systems were in their infancy, I heard a tale (probably a tall tall) where someone was demonstrating a DOS based voice activated commands (yes, that long ago) when some wag in the audience shouted out "format C colon backslash return return" - the tale didn't say if it worked or not...
Alexa places the order - the account holder gets the email, and hits cancel, or sets up the return (or just refuses delivery).
I agree that the universality of the 'start listening' command is a potential issue, and easily resolved, but how big a deal is it to cancel an amazon order - they always take at least an hour or so to actually pack and set up the dispatch (certain addresses in Cambridge not withstanding)...
And the returns process is generally pain free...
Yeah I have an Amazon Echo in my living room every time it hears "Alexa" or sometimes "Alex", "Alexander" spoken on my TV, it lights up and mostly just said "Sorry I don't understand that". Luckily I live in Canada where most of it's US online voice-activated features are crippled. But I've often wondered what that snooping little pod is doing since I know it's listening continuously loading all the sounds it hears into it's memory buffers. When it hears the trigger word "Alexa" it ships the sound hash off to Amazon's Lex API cloud voice recognition service. It's no stretch of the imagination that the NSA could tap into anyone's Amazon Alexa account, or just hack and listen to people's Alexa pods - no need for James Bond to sneak in a wireless microphone. Makes a great bluetooth speaker too!
It has been suggested a few times that you should be able to rename your device.
At the moment the TV/radio only has to not say Siri, Alexa Ok Google.
If you couldn't say ANY name without someone complaining their Olaf had been comparing home and car insurance then we'd be in bigger trouble.
The problem is that the catch words are built-in to conserve precious battery life. So you either have a battery sipper with a fixed vocabulary or a dynamic battery-hog. And no, you can't compare them to voice recognizers of the past that listened for discrete words instead of correcting for context like today's recognizers (which means they couldn't correct for homophones).
"The problem is that the catch words are built-in to conserve precious battery life."
More detail, please. In particular, be careful to distinguish between vocabulary (represented as miniscule quantities of text) and vocabulary (represented as somewhat larger quantities of sound-based stuff, or maybe phone-me based stuff).
If stuff is in real mask ROM, then it's up to the manufacturer to decide in advance what to put in. Sounds a bit limiting in functionality terms (multiple languages?).
If it's in flash memory, then it's in principle alterable, although at some level of inconvenience and energy usage.
If in RAM, then energy usage is highest but so is flexibility.
More detail most welcome.
The problem is that the catch words are built-in to conserve precious battery life.
Thing is.. A 3 syllable word or phrase takes just as much to process as any other 3 syllable word/phrase when you have to process everything even remotely close to tell if that's the target word/phrase. It has to process every thing it hears that could be close enough - and that's assuming there's some level of pre-processing to avoid it trying to process "you're stupid".
However, if I record a phrase in my voice, then it only has to respond to my voice. I can make it what I want (so doesn't have to be a name), and it only has to check what it hears against what it has in memory - reasonable match = listen, otherwise ignore. My much-mentioned T209 was given a 3 or 4 word wake-up phrase that would never come up in ordinary conversation, and that only one other person managed to get past no matter how well he mimicked my voice. And this was mid 90's tech.
Given that the device is going to be sending stuff off over a wireless or other link, processing/sending off audio data and so on, it's rather ingenious to claim "battery life" for such things. Takes more battery life to send audio over WiFi than it takes to compare with a block of internal ram. My T209 had something over 24hours talk time, and enough standby time that it could last a couple of weeks between charges if I didn't talk on it often. So all this waffle about "saving battery life" really is just shite. Especially in an age where most of the western world are used to their phabs needing to be charged every half hour.
Don't forget, the context of this conversation is the wake-up word and letting people change it. Especially away from a fairly common first name, which was a ridiculously stupid idea to use and suggests that Amazon intended such incidents.
Um, the catch phrase DOESN'T get sent to Amazon, only the parts after it, and there are some VERY simple ways to run comparisons to a fixed target (like a ROM, which unlike RAM can still be quick to access AND not need to be constantly refreshed--with RAM, it's one OR the other, not both) using inverse match and delta graphing.
a ROM, which unlike RAM can still be quick to access AND not need to be constantly refreshed--with RAM, it's one OR the other, not both
That's only true of DRAM. SRAM doesn't need refreshing, it's plenty quick enough, and low-enough power that battery-backed SRAM is commonplace.
It's just not very cheap...
Vic.
people buy a voice-activated device which in at least some cases does not tell the voices of its legitimate users apart from everyone else's and which by default can order things from Amazon without additional authentication or even confirmation, place it where it can clearly hear the TV set (which, you know, is known to produce the sounds of human voice from time to time) - and when their device overheard something that triggered it they complained *to the TV station*?
...and I've just given myself the mental image of mainstream broadcast media adding phrases which might trigger Alexa, Siri and the likes to the forbidden-words lists. I think I'll look up how to get in touch with SpaceX to volunteer for a one-way trip to Mars, just in case something like this really does happen.
And on a related note, here is a new fun activity for the more antisocial among the tech-savvy: search the social media for plonkers who have revealed both having an Echo and enough information to allow locating their homes, pop over, stand by an open window and make sure they get all those things a modern home cannot do without, like a 42U server cabinet.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
