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 …
The old, old, old reason to not use entirely-voice interfaces.
"Hey, Siri, I was watching the news and what I heard can't be right.
Siri, can you just delete all my files?"
Whoops.... bye bye files.
Probably that phrasing itself wouldn't work, but saying "Call Mum" near Siri would bypass screen locks and dial your mum for years before it was discovered and an option given to turn it off. My old colleague didn't believe me, so we did it to him.
I have voice in one place - in my car (and I didn't want it but it came "as standard"). But I have to press a button before it starts listening for commands, so it never unwittingly activates and cannot connect online (so the worst that could happen is someone could change a music track or redirect my satnav to a new location).
But Siri, Cortana, "OK Google" (works in raw browsers on many Chrome- or Android-based devices with no special setup), XBox, and now Alexa etc. don't have that. They have a voice command to activate.
Which means a) they are activatable by voice alone, b) they can be activated by accident or pre-recorded message and c) THEY ARE LISTENING ALL THE TIME and trying to recognise what you say. Whether or not they are transmitting that data is besides the point. It's constantly listening out for anything that sounds like a command, recording audio and analysing it. It doesn't take a genius to work out that when they start getting compromises on those devices, you're stuffed and being listened to 24/7 by who-knows, and most likely NOT someone you've agreed a terms of service with.
Voice is a stupid idea.
It's slow, inaccurate, can be activated unintentionally, and cannot distinguish users. You might as well just put a command-line on your sideboard and let anyone type in anything. "Delete all files", "Buy this on Amazon", "Cancel my subscription", "Tweet that my boss is an idiot".
It's game over. Stop doing these stupid things. Nobody needs an always-listening device to order a loo-roll or do a Google search.
You are wrong on "cannot distinguish users"...
This can be fixed with something like Nuance voice print (http://www.nuance.com/for-business/customer-service-solutions/voice-biometrics/vocalpassword/index.htm)
The tech existed 15 years ago should be mature enough to differentiate between you, spouse, tv presenter, and your kids...
Yes, but what about me with a flu or drunk (which ever comes first).
Or both, they're not mutually exclusive :)
I tried the Google thing and I'm pretty sure there's now a record on a US server somewhere that I am daft enough to search for closets in supermarkets :(.
I have shied away from the whole externally processed voice recognition stuff, and not just because I don't want some uncontrolled entity collect data on my habits or keep a perfectly digitised voiceprint on file for later use in some unspecified manner. I already know just how precious little privacy we have left under a triple assault of data thieves like Google criminals and the governments that are supposed to protect us - I am certainly not going to add to that erosion by volunteering to have a listening device in my home.
I love gadgets, but it pays to remain cautious.
We had an African Grey parrot that could mimic everyone in the house to a T and it would have the dogs excitedly milling around the front door with "come on girls, walkies" several times a day. If the dogs couldn't work out it was the bird and not one of us then I doubt that Alexa could either.
My sister gave me an Echo for Christmas and after reading the leaflet I politely suggested she return it as I did not want a spying device in the house. She then asked me to see if my daughter would like it. Daughter's answer was "Ah... hell no! But say it nicely."
The since deceased parrot's worst trick was to respond to a cough with a rolling medley of my parents', sister's and my morning smokers coughs. It got even more gut wrenching after my grandfather spent two months with us while dying of lung cancer and the bird adding his cough to the medley.
my grandfather spent two months with us while dying of lung cancer and the bird adding his cough to the medley.
Parrot intelligence is similar to cat intelligence in its disdain for other lifeforms - except that the parrot is also both smarter and more devious than the cat. The parrot beak is basically engineered for removing keys from remotes and keyboards.
Thus the parrot is the more effective evil, amongst pets.
"I have a voice print recognition to authenticate me on my bank's IVR system. It's been there for several years now."
Right, but voiceprint recognition and "speech to text" don't have a huge amount in common once you start looking in detail. They are different things for different purposes - voiceprints analyse invariant characteristics of the speaker being recognised, sometimes regardless of the words, and speaker-independent speech to text tries to do something quite different - the words are very important, the speaker's identity less so.
"and cannot distinguish users"
Oh, but it can. It can also determine age within certain ranges, gender and a multitude of other things. There are some excellent books available on information to be determined by audio surveillance. People have been better than computers up to now, but the computers are getting very close and even better for some types of info. All you need to do is add a web connected video camera and some system somewhere is going to put the two together and be able to pull up a picture every time it hears a certain voice. 1984 is a little late in coming.
I haven't seen a voice recognition system in 30 years that can get any sentence I say (no matter how simple or deliberately articulated) first time.*
I certainly have never seen one that could use voice as any kind of distinguishing feature between speakers.
The claims are all very nice, but the accuracy still is - and will be for a long time - abysmal.
*Honestly, people test me because they think I must be over-egging it. They bring out their Siri's, their cars, etc. and I say a simple sentence or command that they all understand. I don't have a strong or unusual accent, it's slightly Cockney, that's about it. I can put it in, or take it out of my speech and nothing voice recognition can get even the simplest of commands, in perfect lab conditions.
So it certainly can't be RELIABLY used to tell who I am, it can barely tell what I'm saying and that's without detecting subtleties and nuances of speech and trying to tell me from, say, my brother who - despite the fact that we sound NOTHING alike, everyone confuses us on the phone. Again, people don't believe this, even after meeting us both, and then they ring one of us.
Judging by the school I work in, which has had people try to come in and sell Dragon etc. or library systems, on the basis of voice recognition for writing school reports, or even identifying children for library access (totally non-critical system with humans always present) any number of times and 100+ staff testing them, I'm far from alone.
Voice is NOT anywhere close. In fact, the most impressive voice recog I ever used was bundled with a Sound Galaxy NX Pro ISA card many years ago, along with a speech synthesis software. That got better recognition than ANYTHING I've ever seen or used since (including Dragon, Siri, etc.).
A test:
OK Google, what's the closing time of nearest supermarket?
Just resulted in a blank Google page with the words "slime" and "carpet" in it and nothing else. The room is completely silent except for a cat sitting in the corner licking its bum. The cat probably recognised more of the sentence.
I can't even get "OK Google" to function, though I admit my accent is a little screwed (I'm from Yorkshire, so it's quite deep, but was mainly raised by my Irish nan, so picked up talking at speed, which is fine if you're a high pitched Irish person, but not if you're low pitched Yorkshiremen).
I've similarly had colleagues baffled in disbelief - one colleague in particular who apparently uses it for everything, even dictating text messages in his thick rural Aussie twang (I've emigrated to Sydney) which it gets perfectly, so he brings over his phone to me and says "All you need to say is OK Google" - which it immediately pops up with it's little animated listening screen.
So we cancel it off, wait a good time for it to reset, and I say "OK Google" and..nothing, I try four times, nothing, he's like "But you're saying OK Google clear enough, I can tell that" and sure enough, it pops up ready to search (the I can tell that is a dig as he often jokes he can't make out what I say).
If the future goes to being all voice activated, I shall be left in the past - and given the aforementioned issues and I can generally type faster than I talk, I'm not all that bothered..
"I can generally type faster than I talk".
I wish I could!
A quick google tells me that "The accepted average typing speed is 41 WPM (words per minute), and professional career typists can exceed 100 WPM"
Also: "..average American English speaker engaged in a friendly conversation speaks at a rate of approximately 110–150 wpm."
I can't even get "OK Google" to function, though I admit my accent is a little screwed (I'm from Yorkshire, so it's quite deep, but was mainly raised by my Irish nan, so picked up talking at speed, which is fine if you're a high pitched Irish person, but not if you're low pitched Yorkshiremen).
Why am I now imagining Compo from "Last of the Summer Wine" walking up to one of these things and saying "ey'oop Google !" ?
"But you're saying OK Google clear enough"
Does your buddy not remember setting up "OK Google" where he has to repeat the phrase several times so that it will learn his voice so that it only responds to him? Obviously he forgot that...
I dont doubt that it cant handle your accent, voice recognition struggles with mine too (An odd mix of Norfolk, Lancashire, Central Scotland and the highlands) but I generally find speaking slowly and loudly works (pretend you're on holiday in Spain).
Perhaps you should have said "the nearest". I have met quite a few Americans who complain about having difficulties in understanding the English some Brits produce. One could suppose those programs are made and tested in the USA. I am not impressed by this whole voice recognition thing but it's obvious it's getting better and that there are individuals who will welcome it, like say blind persons etc. Text to voice has also improved while still awful. The Joke Alert for the "the".
If Alexa asked for confirmation, and they showed it on TV, it would get the confirmation.
The bug is that everyone addresses their assistant the same way. They are all Alexa, they are all Siri, they are all OK Google. The first thing they should do is ask you to give them a new name that can't be their actual name.
Of course, if you let your little girl name the Alexa, they will all be called Elsa, and we're back where we started...
The first thing they should do is ask you to give them a new name that can't be their actual name.
My ancient Sony Ericson T209(?) effectively had that. You recorded a word for it to listen for (eg name) and after that it tried to listen for further commands. It didn't come with a pre-set name, you had to put it in yourself.
"If Alexa asked for confirmation, and they showed it on TV, it would get the confirmation."..
Maybe the "confirmation" phrase should be unique/change per each command:
Poor, drunk sap: "Alexa, please purchase 10 Tesla Model S vehicles"
Alexa: "Please confirm by saying "%)&@()*&$)@&$@)&$*@&)$#(&@)(#&$)(@*&$"
Poor, drunk sap: "arrrrrrrggggggg"
Alexa: "Would you like a different confirmation phrase?"
That's easy, you should have two language settings - one for English, one for American.
Why not? They are in reality rather different versions of English, sufficiently so that RDBMSes have to have options for them. Years ago we had to stop DBAs from setting the language for our software to British on SQL Server, which promptly stopped our (International) date format from working.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
