Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Phrases that incorrectly trigger Alexa, Siri, and Google Assistant (arstechnica.com)
90 points by BerislavLopac on July 1, 2020 | hide | past | favorite | 45 comments


Might try switching to "ok, cool" for Google Assistant since "hey google" is one of the most awkward/ugly sounding phrases to have to repeat over and over.


After saying “hey sweetie” to my baby girl and triggering Siri for the thousandth time, I'm _wishing_ for “Ok Google”…


Just tried it now and the intonation seems important. If I say ok cool normally there's no trigger. But if I say it with the same intonation I would say ok Google then it triggers.


Someone on the Alexa team told me that they had false triggers from the phrase 'put your pajamas on" sounding like Amazon.


I love this.


My partner and I speak Chinese and English at home and I cannot tell you how many times random Chinese phrases trigger my Homepod (set to listen for English).


The trigger phrase for Mandarin Siri is “嘿 Siri”, so anything involving black something would most likely trigger it.


I'm surprised at how sensitive they can be. I was listening to "Wait Wait... Don't Tell Me!" on NPR one morning in my living room, at a fairly low volume.

I noticed that there was music coming from my kitchen. My Google Home Mini that serves as a kitchen timer was playing a Frank Sinatra song. According to the logs it heard someone say "Hey Google, play some Sinatra".

I was surprised that it would hear a low volume radio in the living room from the kitchen. Also, even if it thought it heard "Hey Google", there was nothing that should have sounded like "play some Sinatra".

The Echo Dot in my living room puzzles me in the opposite direction. Amazon runs TV commercials advertising Alexa and Amazon Music that consist of some song by a particular artist playing over some animation of a singing mouth, then the music stops and the mouth says "Alexa, play <artist> on Amazon Music".

That frequently triggers the Dot, but only once has it actually then played music in response. Here we have it hearing a clear "Alexa", triggering, and somehow realizing that it should not actually do the command. How does it know?


Apparently, there's a regularly updated local datastore with signatures of currently running commercials with wake words in them: https://venturebeat.com/2019/01/31/why-alexa-usually-wont-re...


Just for giggles I call it Alissa. It works most of the time, other times it lights up and goes quiet again.


The one that triggers Siri most often in my home is "Are you serious?". My wife says that to me all the time.


"Seriously" is the word which constantly triggers Siri for me. I feel your pain.


I've heard that Google WFH during pandemic has lead to funny Google Assistant issues.

Since many Googlers have on in their kitchen, mentioning 'Google' on a call often triggers a chain reaction.


For Google devices, in the Home app under Device settings → Accessibility, set ‘Start sound on, end sound on’. Then you know exactly when it's active.

The examples in the article don't work for me, but “Hey Booboo” (as in picnic basket) is pretty reliable.


If you want to collect phrases that trigger Google, just drive around listening to books on Audible through Android Auto.


What if it activates while you’re units-distance-away and don’t hear the start / end sounds, for whatever reason.


Yes, first thing I do with every new smart device. This should be on by default.


Side note: One side effect of covid I've been thinking about is the fact that usage of these voice assistants will grow from people being at home, and alone, a lot more.

I never even thought about using one before the virus, primarily due to the fact I was usually around people, and it was either impractical or embarrassing to ask Alexa, Siri, or Google to do something for me. But now, like millions of others, I'm stuck at home all day and it's actually quite natural to take advantage of the voice features.


Out of curiosity, what do you use it for that you didn't before? I've personally found voice assistants to have an extremely limited set of use cases that are actually faster or more convenient than pulling out my phone.


Quick conversions in the kitchen ("what's 1/3 cup to tablespoons?"), controlling smart home features ("turn on the living room lights", "tell the thermostat I'm hot"), getting weather forecasts ("what's the weather today?"), etc. I've found the use cases increase the longer you have one.


Sending texts, setting timers, making calls, etc. Nothing earth shattering, but it's growing on me.


I use it to put some music or listen to the news without having to pull my cellphone and risk getting sucked into Reddit.


So if I go down the IOT aisle at Best Buy and say

"Ok, who is reading a city in Montana a letter?"

everything will light up?


Especially coming from spending years in the ASR industry and having several Alexa devices at home, I find the false-positive rate remarkably low given how high the accuracy is for the wake-word and server-side reco, even from quite far away. (The whisper-to/whisper-back feature is also freaky, and actually found myself using it the other night so as to not wake up my son.)


> the false-positive rate remarkably low given how high the accuracy is for the wake-word

This was a major focus of the speech science team... product team and leadership identified very early that the bias had to be in this direction to build trust. Also the reason there is a visual signal and an (optional) audio signal for "wake up" and a mics-off button.


"Cocaine Noodle" can reliably trigger Google Home from my test.


And still less awkward than "Hey Google".


There are electronic timers with one-button activation for a fixed period of time (15/30/60/120 mins), which can be used to avoid voice assistants listening 7/24.


I attended a Passover last year and when the some of the people used the word Elijah, an Alexa in the room would respond. They were all from NYC, FWIW.

I’d never heard of “Elijah” so asked the Alexa for an explanation, which we got. Followed by some work-related notifications for the host. (Luckily nothing confidential was uttered, it seems)


"Hey, Sarah" works around 80% of the time. Not surprising. Still a pain when your wife is named Sarah.


I set all my Alexa devices to wake to "echo" first thing. Echo Auto has been rendered useless to me, because (last I checked) it doesn't allow changing the wake word. It's impossible to listen to radio without triggering "alexa' during every single commercial break.


Seems to me Alexa got it better than any other. It's easy to say and not awkward like "Hey Google", and I can speak my command in a flow without pausing. Shame that the AI backing it is quite possibly the worst among its contemporaries.


Our cat's name "KJ" triggers "Hey Siri" frequently, usually when my wife says it. Seems like such an odd cross connect.


"In theory" triggers Siri for me often.


In your accent, do you pronounce it as "in thee-uh-ree/thee-oh-ree" or "in thee-ree?" If the later the vowels might be overlapping with the vowels in siri?


More "in thee-ree." I'm not surprised at all by it, as for me, "Hey, Sirir" is definitely close to "In theory."


I turned off Google Assistant for this reason. I say "ok cool" often enough that it became a source of frustration.


This happened to me just yesterday when I said to someone, “are you serious?“ And Siri came on and asked me for follow up.


I say “are you serious” probably way too often and it almost always triggers Siri.


We named our dog Sami. HomePod triggers probably every 2-3 days.

A shower triggers the Apple Watch Hey Siri probably once a week.


Somehow Playstation 4 has flown under the radar, it is the most finicky!

I should probably turn its voice control off


Si, Si, is apparently number 1 in Spanish speaking communities for incorrectly triggering Siri.


Ok boomer works on the GH too


[flagged]


This comment doesn't add anything to the discussion. Nobody has mentioned race in the context of this article and, although there are well documented racial biases in both human and machine profiling systems, it's irrelevant here.


Ars jumps the shark. Uncovered is the word used in this headline? Really? Good lord who are the losers that go into tech journalism. How is there any money left to pay these click baiters?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: