The fact that these voice response devices don't lock into a particular set of users' voices is a major problem. In fact, multiple times I've had a Google ad cause the phone it was playing on go do a web search triggered by that ad. Totally insane.
My wife and I read my four year old son story books at night before he goes to sleep. And multiple times while I am reading the story, Siri on my wife's iPhone has interpreted my voice as a command to do a web search for something–something totally unrelated to the story I am reading. Strangely, only my voice has ever triggered Siri in this way, when she reads the story Siri ignores her. (Not on my iPhone, first thing I did when I got it was turn Siri off.)
The models listening for the initial hook of "okay google" or "hey cortana" etc are stored and executed locally on the device-- usually in a low power manner.
This comes with a set of constraints:
Certain activation phrases are more catchy than others, with hard consonant sounds and enough length for the alg to grab on to. Allowing users to pick any name opens this up to a high false positive rate, where the device would 'respond' to arbitrary inputs rather than specifically its name when called.
Training models on new trigger words and then deploying them back down to the device is in its infancy, and voice detection has a long way to go (another example of this is the difficulty around responding to a specific person's voice, with its signature sounds etc). I could see this happening in the future as the tech improves.
My understanding is that these devices have ASICs which listen for their names, the actual pattern for the name coded into hardware, meaning that you can't easily select a new name.
Voice recognition takes more CPU power than the devices have, especially if you are trying to prolong battery life, and I think the more complex commands are relayed to a server instead of parsed locally.
Low-power DSP chip, I believe. For more complex tasks like voice recognition it generally makes sense to use something that at least resembles a general-purpose CPU.
Siri isn't triggered by TV in the same way. Take a look at last week's keynote where Phil Schiller says "Hey Siri" on stage and takes a pause to note nobody's phone was activated by his voice.
I wonder if there is a low-pass filter on Alexa's microphone? Triggering your Alexa with ultrasonic commands outside of normal human hearing would be even more entertaining. Ads already embed ultrasonic chirps to some extent for tracking purposes.
These creators are just great guys. Not only is this again a sign of them being on the edge of healthy creativity, but they're also the only show that just has all episodes for free to watch.
There's a documentary on how they make each show "a la carte" each week, and never missed its deadline except once, but it's often been a close call of just a few minutes.
I thought the only reasons they offered their content for free (ad supported) was that they were competing with a major piracy site dedicated exlusively to streaming their episodes.
I read years ago that they explained that they placed the episodes free online because they were tired of going to pirate sites to get convenient access to their own stuff.
Audio is a terribly insecure channel to accept commands through. You presume no one else has access to the device because it is physically locked in your house, but sound travels through walls and through speakers. Heck, you can even make a window pane into a speaker with some simple gear.
It's no big deal when the worst case scenario is being served up a search result you don't need or listening to a song you didn't request, but you definitely wouldn't want to link voice recognition tech up to anything non-reversible like a stock trade or the lock on your front door.
> DolphinAttack ... modulates voice commands on ultrasonic
carriers ... to achieve inaudibility. By leveraging
the nonlinearity of the microphone circuits the modulated ... commands can be successfully demodulated
Yeah, but it's the way it was done in Star Trek, and many of the engineers creating the technological future want that future to behave like Star Trek. Voice command isn't a technically superior user interface, it's just more "futuristic."
And <tangential rant> it shouldn't be mandatory. If I want Cortana or Siri or Alexa, let me download and install it. Don't FTLOG integrate it so deeply into an operating system that it's all but impossible to avoid </tangential rant>
Tim Cook's voice (or someone else from the presentation) enabled Siri on my phone during the iPhone X event (watching over YouTube). I thought it was supposed to be locked to my own voice though?
Indeed - but it's a much more difficult problem than it seems at first glance (unless you're okay with an always on, always streaming to the cloud microphone).
47 comments
[ 3.2 ms ] story [ 75.5 ms ] threadI still don't understand why these companies cannot let us choose a name for our listening devices. Computers can be named, phones too.
This comes with a set of constraints:
Certain activation phrases are more catchy than others, with hard consonant sounds and enough length for the alg to grab on to. Allowing users to pick any name opens this up to a high false positive rate, where the device would 'respond' to arbitrary inputs rather than specifically its name when called.
Training models on new trigger words and then deploying them back down to the device is in its infancy, and voice detection has a long way to go (another example of this is the difficulty around responding to a specific person's voice, with its signature sounds etc). I could see this happening in the future as the tech improves.
Voice recognition takes more CPU power than the devices have, especially if you are trying to prolong battery life, and I think the more complex commands are relayed to a server instead of parsed locally.
Demo: https://www.youtube.com/watch?v=21HjF4A3WE4
Paper: https://arxiv.org/pdf/1708.09537.pdf
https://news.ycombinator.com/item?id=15199233
There's a documentary on how they make each show "a la carte" each week, and never missed its deadline except once, but it's often been a close call of just a few minutes.
Edit: "6 Days to Air: The Making of South Park"
https://www.youtube.com/watch?v=gQ-Un8JjUwo
They haven't been free for years.
http://southpark.cc.com/full-episodes/s13e04-eat-pray-queef#...
Perhaps they're blocked in the US?
It's no big deal when the worst case scenario is being served up a search result you don't need or listening to a song you didn't request, but you definitely wouldn't want to link voice recognition tech up to anything non-reversible like a stock trade or the lock on your front door.
From https://arxiv.org/pdf/1708.09537.pdf
> DolphinAttack ... modulates voice commands on ultrasonic carriers ... to achieve inaudibility. By leveraging the nonlinearity of the microphone circuits the modulated ... commands can be successfully demodulated
And <tangential rant> it shouldn't be mandatory. If I want Cortana or Siri or Alexa, let me download and install it. Don't FTLOG integrate it so deeply into an operating system that it's all but impossible to avoid </tangential rant>
Other comments indicate the assistant names are hard coded in ASICs but they certainly don't have to be.
https://www.amazon.com/GOMPHOCARPUS-PHYSPCARPUS-PLANT-HAIRY-...