We got an Echo Dot for Christmas and have enjoyed using it.
A few weeks after we hooked it up, I watched a crappy made-for-TV movie I’d recorded. A minor character’s name was Alexis. Several times the Echo took a character saying its name as a question, so I’d get one of those, “Sorry, I didn’t understand the question” responses.
During March Madness, Amazon ran some commercials for Echo in which someone would call on Alexis. Our Dot never once responded. (I think, but am not 100% sure, that at least once we watched the commercial rom our DVR an hour or more after it aired.)
Last week we watched an episode of The Daily Show we had recorded. They had a segment about a toddler that asked Alexis to play a game, which was amusingly misinterpreted as being a request for porn. (I did wonder if this was a set-up of some sort.) But our Dot did not respond to the child’s request.
So how does Echo know when it’s actually being called on? I find it remarkably accurate at detecting requests; it occasionally responds to something that wasn’t its name and even more occasionally doesn’t detect when it has been called on, but these instances are rare.
I despise computer voice interfaces and have banished them from my life, but I thought a key feature was that you could change the voice activation phrase. It doesn’t have to be “Alexis”, “Siri”, or “OK Google”, it could be “HAL 9000”, or “Computer Wife, I am needy” or something completely different, right? Or was that all tossed for branding purposes?
The Echo/Dot learn over time how to recognize your voice. It will still respond to anyone but it gets better at filtering out false positives with use.
Nope. They will not let you change the activation phrase for some reason (I guess to avoid people tarnishing the brand with off-color names/words). That said the Amazon device will respond to a few different wake words: Alexa, Echo, Amazon and Computer.
Not sure about the Google device but I suspect it is similar.
I can attest that the Google device in my house recognizes “Hey Google” and “OK, Google” when spoken by a human in the house. It also recognizes it when commercials on the TV say those phrases (it’s about 50/50 on responding appropriately to commercials, often becoming confused by background noise).
And while I seldom hear it say it’s version of “I’m sorry, I don’t understand” in response to random noises, I frequently hear it making the noise I refer to as its “ears pricking up” in response to seemingly random words on the TV or radio.
There was an episode of the last man on earth where Melissa was running around, and Gail said something to her and that phrase activated my Echo. I even rewinded hulu to watch it a couple more times and each time it activated the Echo. I guess what Gail said sounded like “Alexa” despite the fact that she was saying 'Melissa".
So I’ve noticed some youtube videos will activate my Echo, but TV commercials will not. I’m not sure why. Do all the echos get information from Amazon telling them to ignore TV commercials somehow?
I was trying to connect my phone to Alexa but didn’t know how. I looked up a youtube video that showed how to do it, and the instructions in the video woke up my Alexa and connected it to my phone automatically. I still don’t know how to do it
More likely than not, a machine learning algorithm is being used to recognize the word “alexa” and it is just continuously being trained given the constant stream of positive and negative “hits” it gets. Unfortunately for us, that doesn’t provide an easily observable answer to “what rules does it use?”
I’ve had my Echo go off a few times when I’m almost positive nothing close to “Alexa” or “Echo” was said.
What’s really weird is that it seems to really need to be spoken to very specifically - “Alexa! - beat - play…” When my nieces just talk to it “Alexa what time is it…” it doesn’t understand.
So I think it’s weird that random voices from the tv can get it to respond
Maybe it also listens for intonation. If the name is said as part of the sentence, it doesn’t pick it out but if you call or question it, it does. Or maybe just the speed, clarity and pausing that you use when you are getting someone’s (something’s) attention helps it to hear its name.
Here is an article on specific machine learning algorithms used for speech recognition.
Architecturally, there is probably a very simple neural network being run directly on the Echo box that does nothing but recognize the few activation phrases, and then transmits whatever comes afterwards to Amazon for more sophisticated processing.
These devices can block known irrelevant phrases. Burger King has run some commercials where it uses a phrase to cause Google devices to start reading the BK Wikipedia page. Google soon blocked the phrase from the ad. BK then modified the phrase, etc.
It’s not really “blocking the phrase”. It is recognizing the specific recitation of that phrase from the commercial and blocking that. Since the commercial says it the exact same way every time it is far easier to train the algorithm to recognize the ad.
Oddly, the ones I’ve encountered do respond to “Alexis”, in that the blue ring lights up and it appears to listen to you. But then, it does nothing. No action on your instruction. Try it yourself, first ask Alexis to do something, and then ask Alexa the same thing.
