When it comes to intelligent voice assistants, everyone will definitely think of those experiences that are speechless by them.
After all, at this stage, artificial intellectual disabilities are really far from intelligence.
Don't be afraid to laugh at everyone, Shichao really used Siri to improve his life efficiency a long time ago.
The simplest thing is to let him check the real-time exchange rate, identify the song, and set a time.
A little more advanced, with the Mijia App and iOS shortcuts, just say "Hey Siri, I'm home" after work.
The curtains in the home will be closed, the mood lights will be turned on, and the Bluetooth speaker will be activated together.
Eh, that sounds pretty cool, doesn't it?
But after practicing it countless times and finding that Siri's listening and comprehension skills sometimes really wanted to be forced to give it two big fights, I gave up.
There was that time of repeated correction, and my TM fingers could poke well.
So like many iPhone users, Shichao now uses Siri to set a time, and others really don't expect this to be smart.
However, some time ago we sent an article on network security, and the message at the bottom made me have a little concern about this big smart.
—— When we have Hey Siri, the phone has a response, so when it is not hi, does the phone not always listen to the sounds around us?
Sizzle, seems to make some sense?
If you don't keep listening to the sound, how does it know I'm calling him?
Later, I touched apple's official website and found that they mentioned four words in the Siri introduction copy, "always on call"...
My suspicions grew heavier.
If it is really like the netizens said, the mobile phone has been listening to us, then is it not the midnight reading of the study materials that are known every day?
With this question, Shi Chao searched the Internet all day and really found some claims.
To conclude, iPhone can hear you shout Hey Siri, indeed because it has been listening to your voice.
But don't be in a hurry to squirt, he can understand the sound ≠.
It's like your dog can hear you all the time, but can he understand it all?
To figure this out, we'll have to move back in time to 2013.
That year, Apple released the iPhone 5s, in addition to cramming the iPhone with the latest A7 processor, but also adding a small M7 coprocessor.
The A-series processors that everyone is familiar with are the C bits in the iPhone, which are responsible for most of the work.
The coprocessor is the equivalent of an auxiliary, responsible for some things that are not difficult, but need to work all the time.
For example, the number of steps of the PK in your usual WeChat is the acceleration sensor and gyroscope perception, which is handed over to the coprocessor for calculation.
The reason is simple.
Because the power consumption of the coprocessor is very low, even if the data is calculated all the time, it has no impact on the battery life.
If this is done for the A-series processor, the mobile phone is equivalent to waking up all the time, and the battery life has collapsed early TM.
In addition, the compass and the pressure sensor are also coprocessor responsible.
After Apple introduced the Hey Siri feature in 2014, the coprocessor had one more task: processing Siri audio.
The microphone collects the sound and the coprocessor recognizes it.
So the phone really has been listening to the sounds around you to judge if you're saying Hey Siri.
The question is, can the mobile phone understand the surrounding sounds at this time?
Hard.
Let's start with Siri's recognition principle.
When talking to Siri, the sounds collected by your phone are sent to Apple servers for calculation, so it's easy to know what you say.
But before triggering Siri, the phone has been collecting sound, the data is very large, if tens of millions of iPhone data are sent to Apple, the server will have exploded.
So before triggering Siri, the sound collected by the phone is actually calculated locally.
This calculation is very complex, you can refer to the following figure.
Because Shichao is not a related major, this figure can only be understood for a small part, probably to explain to everyone.
Once your voice enters the microphone, it is processed to cut into many small pieces of audio, and then the coprocessor calculates the "score" of this audio.
Finally, add up these audio scores and see if you reach your "Hey Siri" sound threshold.
Once reached, the coprocessor wakes up the main processor, which is determined by a more precise set of algorithms.
Confirm that it's you shouting "Hey Siri" before Siri comes out to get the bell.
To summarize:
After the coprocessor continuously detects the surrounding sounds and confirms that it is "Hey Siri", the main processor counts again and requests the Siri server.
After such a complex set of processes, the iPhone can determine that it is hearing "Hey Siri", you say that it has to understand all other words, then the workload is not enough to double a hundred times and a thousand times.
So, it's not realistic that Siri can understand you before it's triggered.
Is it possible that, whether you understand it or not, it just collects the sound and passes it directly to Apple?
Technically, it is feasible, but in terms of battery life, it is not very feasible.
You think, if the mobile phone has been transmitting data to the outside, the battery is bigger than your head, and the battery life can't stand up, right?
Even if Apple is sinister enough to only secretly transmit it when charging, such a large data traffic is not open to send people's heads.
When you see this, you will say, are you washing apples?
There has been a foreign news before that Apple has heard an indescribable voice through Siri?
That I also went to check, the article is half a title party, the actual situation is not Siri eavesdropping, but the user mistakenly triggered Siri, Siri began to listen, at this time the user is doing something shameful.
In order to determine whether Siri is working well, Apple will extract less than 0.2% of the audio information and give it to the human to view, and finally crash into it.
Users can also refuse to share audio ▼
Of course, in addition to Siri, there are other intelligent voice assistants, most of which also listen to wake words through a low-power chip.
It can be said that these intelligent voice assistants (including Siri) can understand the surrounding sounds or upload all the sounds to the server before triggering, whether it is from the technical level or from the logical judgment, it is not very realistic.
Is it possible that under certain conditions, the voice assistant will quietly connect to the Internet and eavesdrop for more than ten seconds?
It is not often said that the front foot just talked about a product, and the back foot it appeared on the online shopping platform, suspecting that it was a voice assistant or something to eavesdrop on the chat.
In fact, Shichao feels that this matter has a high probability of relying on the advertising alliance. They have a large number of accurate portraits of users, such as who likes the second dimension, and have recently been looking at daily products.
According to the user portrait to recommend products are really accurate, sometimes just happen to the front foot said, the back foot hit the recommendation, let people mistakenly think that they are not eavesdropped.
We've also written about ad networks before▼ ▼
Of course, it's not that voice assistants won't eavesdrop, there are possibilities, but it's relatively small.
And taking a step back, with the implementation of the relevant privacy laws on the mainland in the past two years, major manufacturers have stepped up authority management, and I think the possibility of eavesdropping is lower.
If you think about it, at the risk of being sanctioned, or even breaking the law, eavesdropping on a few words from the people, just to make a profit, for any voice assistant manufacturer, it is not necessary.