PSA: If You’re Late To Work, Try Blaming The Pixies
from the stop dept
I’m a weird guy, but you all knew that already. One of the ways in which I’m particularly weird is that I have not used any sort of an alarm clock in roughly 20 years (I’m in my 40s). For whatever reason, I can decide in my mind when I want to wake up and then it just sort of happens. All of which is to say that this story about an alarm clock app is largely foreign to me.
But if I did use one, and if I ever found myself running late to something as a result of not waking up on time, I would immediately blame famed-band The Pixies, and you should, too. That is because one the band’s hit songs appears to successfully stop Google’s Assistant alarms on several smart displays and phones. How? Well…
It wasn’t long ago when Google rolled out a handy feature that made Assistant alarms shut up by simply saying “Stop” or “Snooze” — none of that “Hey Google” nonsense needed. These quick phrases were first available for alarms and timers on Google Assistant smart speakers and displays like the Nest Hub, but they eventually made their way to Google’s phones starting with the Pixel 6. Now, users are reporting some hilarious unintended side effects with this feature.
After noticing their alarms didn’t seem to go off on certain days, one Redditor woke up early to get to the bottom of the issue. The Pixies song Where Is My Mind? started playing since it was in a Spotify playlist the user had set as their music alarm in the Google Clock app. If you’re familiar with the song, it starts with a person singing “Oooh,” then someone else cuts them off by exclaiming “Stop!” before music starts playing.
And you can already see where this is going. The app hears the command at the beginning of the song and complies. Seconds into the alarm going off, it ceases because the band’s Kim Deal tells it to.
But per the source post, this isn’t as wide-spread a problem as you might think. The folks over at Android Police and others tested a variety of songs that have the word “stop” in them, such as U Can’t Touch This and Ice Ice Baby, but those did not have the same effect. Why? Likely because Quick Phrases is set to ignore those words if they are part of a song, which is what makes the Pixies song somewhat unique.
Where Is My Mind is unique in the fact that its “Stop!” comes before music really starts playing. Google’s Quick Phrases feature seems to ignore the command when it’s backed by instrumentals or vocals, even when the music drops out as the phrase is said. That lines up with all the user reports we’ve seen complaining that Assistant never hears them yelling “Stop!” over the sound of the music playing when they’re actually trying to use this feature.
It’s definitely an edge case, and a funny one at that. But Google might do well to teach its word detection model to ignore this one sound bite in particular.
Or the world can learn to magically command themselves to get up at a certain time like I do. Though, I’ll admit that might not be entirely feasible.
Filed Under: google alarm, google assistant, stop, the pixies, voice recognition
Comments on “PSA: If You’re Late To Work, Try Blaming The Pixies”
In the name of love!
I can’t help but wonder if anyone’s tested Stop In The Name Of Love by the Supremes, or if we’re getting eccentric, Pink Floyd’s Waiting For the Worms in which Pink (Roger Waters) screams STOP! at the very end.
Yeah they meant for Kim to stop because that was all the sample they needed. But now your alarm app shall obey.
Shouldn’t Google devices be ignoring all sounds that they’re producing themselves? Yes, this instance is funny, but what if a person were watching a video in which someone yells “call 9-1-1!”? (Particularly if the character is actually yelling it to a Google voice assistant, with the proper preamble.)
This is kind of like echo cancellation, but with much a looser real-time constraint: run voice recognition on everything being sent out the speaker(s), and if that exact thing is heard over the microphone within a hundred milliseconds, ignore it.
Signal Processing 101
Lesson 1: When analysing an incoming signal, the first step is to filter out anything resembling the outgoing signal.
Addendum: Filtering out the ill-begotten remnants of echoes that bounce back from walls and furniture with their phases all banged up is kinda hard.
Re:
Luckily, it’s not necessary. Such quiet remnants of sound are annoying with speakerphones or public-address systems, especially when they cause feedback. But nobody’s gonna hear the badly-filtered result in this case. Just ignore sufficiently quiet or “echoey” sounds (maybe updating calibration data continuously). People usually won’t want these things responding to commands from outside an open window or across the house anyway.
Re:
That’s standard echo cancellation fare that smartphones do pretty well.
The people waking themselves up with playlist items at home are more likely than not using some Bluetooth-connected loudspeaker contraption instead of the tinny smartphone speakers. That adds significant amounts of processing and delay, making it quite harder to capture in echo cancellation than the comparatively controlled connection from its own speaker to the microphone.
In addition, good echo cancellation needs a training signal to work with. Unless the phone precedes its first utterance with some seconds of subliminal noise or pseudonoise, its echo cancellation will start out deficient. When using free-hand telephony, this is masked by switching the audio at first to half-duplex: if something is output to the speaker, the microphone is muted completely for a brief period until echo cancellation is suitably trained.
Now completely ignoring a user shouting “Stop!” or other commands for a few seconds when the smartphone suddenly starts blaring comes with its own acceptance problems…
In short: the maxim “smartphone voice recognition should ignore anything originating from itself” is certainly a sensible paradigm, but its implementation may end up trickier than its statement.
This illustrates two things:
First, the evil of voice command. Picture being able to shout “are em dash are star dot star enter” (rm -r .) near your co-worker’s cubicle.
Second, that Google Assistant has an even bigger problem, that they tried monkey patching their way around: that commands from the device are still being accepted by the device.
If you think The Pixies are bad, picture some well-created malware that plays an audio clip directing google assistant to ship money out somewhere. Perhaps directly by banking, perhaps through purchase of a grifting product, perhaps by helping install OTHER malware. And because the audio sensitivity of the device is greater than the human ear, it can be played at a volume lower than you can hear, or at frequencies outside your range.
Yay, automated assistants! Written without considering the active malevolence of humanity.
Re: shutdown, yes, yes
When the Mac 2 came out with voice recognition built into it’s OS, running into the computer lab and shouting “Shutdown, yes, yes!” Was a common thing to do. Everyone would lose all work since the last save.
Those who will not learn from history are bound to repeat it.
This is one of the reasons I disable the assistant.
The only place I allow it to be active is while Android Auto is active, since it’s supposed to safer to interact with while drive, though tbh I don’t use it much in the car either.
The alarm app is bad enough with Google constantly changing the button labels/icons and press vs slide behavior.
When you’re drowsy and waking up and the screen is blurry, it’s hard to see which button is snooze, and which is stop.
I already press Stop accidentally often enough due to that, I don’t need my alarm to stop randomly for some other reason. Or worse, as someone else mentioned, having it dial 911 for me with or without my knowledge.
When your AI starts muttering to itself and hearing voices:
RUN!
OG Pixies!
If you’re that into OG Pixies, you’re awesome and you deserve more sleep. Problem solved.
Fullstack Developer
Full-stack developers are in demand everywhere, considering the wealth of frontend and backend development skills they bring to the table. Pune is one of the most important IT hubs in the country, many budding IT professionals are searching for a competent and career-oriented full-stack web development course in Pune and going for full-stack training in Pune.
I stopped reading after the 3rd sentence. Timothy, you are Cozmo Kramer (but with employment).