There are references to different research studies, one look at inaudible commands, and the other to embedded into other sounds or noise. I think the inaudible is easier to avoid, unless it uses effects of lower harmonics reacting to the actual body where the mic sits, to be able to produce local vibrations in the human voice band.
One of the most useful comments I’ve read in a long time, if not ever on MacRumors! Thanks for this!Consequence of not using a proper analog anti-aliasing filter before the ADC? If the signal is not analog lowpass filtered to half the sample rate before sampling then those frequencies above half the sample rate 'alias' down into frequencies within the normal sample range. For example, if the sample rate is 40khz then sounds between 20khz and 40khz (outside the normal hearing range) sound identical to the sampler as sounds between 0hz and 20khz do (inside the normal hearing range). This is the audio equivalent to those Moire patterns you sometimes see in photos with textures finer then the pixel density of the camera or monitor.
If so this would imply they tried saving a few cents by either leaving out the front end analog lowpass filter or using a very cheap first-order filter which isn't something that could be fixed in firmware. A cheap filter can be tricked by just increasing the volume of the sound and since you can't hear it anyway it doesn't matter how loud they make it. This must be the approach they used since the article states "While DolphinAttack has its limitations — the transmitter must be close to the receiving device — experts warned that more powerful ultrasonic systems were possible."
Fixing this requires designing in stronger high-order front-end filtering which ultimately should only help Siri actually recognize you when you 'do' speak. And we already know she needs all the help she can get
Here is a white paper describing the phenomenon...
http://www.ni.com/white-paper/54448/en/
Sure, but human voice frequencies don't necessarily match human hearing frequencies. Meaning the human voice could quite possibly produce frequencies that human ears cannot hear. Even if evolution will have tended to get rid of those non-useful frequencies (though who knows, some of those frequencies might have useful when interacting with animals), some might have been impractical to get rid of (ie, human vocal cords optimised for human hearing might still produce other frequencies as a by-product). Plus there is the whole ageing aspect in that older people cannot hear higher frequencies.
Sure, because all those people with voices outside the human hearing range can communicate so well with other people already.
Speech recognition is one step, but I got a Siri response with:Wait, so if I get a dog whistle, Siri will finally understand me?
Balancing my bike at the traffic lights, looking cool AF triple tap AirPod "Set volume to 80%"
..."William Oliver Stone (born September 15, 1946) is an American writer and filmmaker"
Sometimes I think Siri just wants me to die single
That was a thing? Oh man, maybe my ears were always old. Also, that sounds like torture.Reminds me of the mosquito ringtones that kids used for a while to hide calls from older folk
As some have already noted, this should prompt makers to filter input to voice frequencies. Easy.
![]()
Researchers in the United States and China have been performing tests in an effort to demonstrate that "hidden" commands, or those undetectable to human ears, can reach AI assistants like Siri and force them to perform actions their owners never intended. The research was highlighted in a piece today by The New York Times, suggesting that these subliminal commands can dial phone numbers, open websites, and more potentially malicious actions if placed in the wrong hands.
A group of students from the University of California, Berkeley and Georgetown University published a research paper this month, stating that they could embed commands into music recordings or spoken text. When played near an Amazon Echo or Apple iPhone, a person would just hear the song or someone speaking, while Siri and Alexa "might hear an instruction to add something to your shopping list." Or, more dangerous, unlock doors, wire money from your bank, and purchase items online.
![siri-iphone-x.jpg]()
The method by which the students were able to accomplish the hidden commands shouldn't be a concern for the public at large, but one of the paper's authors, Nicholas Carlini, believes malicious parties could already be making inroads with similar technology.
Last year, researchers based at Princeton University and Zheijiang University in China performed similar tests, demonstrating that AI assistants could be activated through frequencies not heard by humans. In a technique dubbed "DolphinAttack," the researchers built a transmitter to send the hidden command that dialed a specific phone number, while other tests took pictures and sent text messages. DolphinAttack is said to be limited in terms of range, however, since it "must be close to the receiving device."
In yet another set of research, a group at the University of Illinois at Urbana-Champaign proved this range limitation could be increased, showing off commands received from 25 feet away. For the most recent group of researchers from Berkeley, Carlini told The New York Times that he was "confident" his team would soon be able to deliver successful commands "against any smart device system on the market." He said the group wants to prove to companies that this flaw is a potential problem, "and then hope that other people will say, 'O.K. this is possible, now let's try and fix it.'"
For security purposes, Apple is stringent with certain HomeKit-related Siri commands, locking them behind device passcodes whenever users have passcodes enabled. For example, if you want to unlock your front door with a connected smart lock, you can ask Siri to do so, but you'll have to enter your passcode on an iPhone or iPad after issuing the command. The HomePod, on the other hand, purposefully lacks this functionality.
Article Link: Researchers Demonstrate Subliminal Smart Device Commands That Have Potential for Malicious Attacks
Would it be possible for them to split the signal path? First using a high-pass filter that only looks at the ultrasound range and if the signal contains indoor GPS information process as such? And then using a low-pass filter to only look at the range below ultrasound to process for any voice commands?
