There is so much effort and thought behind the phrase “Hey Siri”

Apple today published a new article on the company’s machine learning blog, which describes how hardware, software, and Internet services come together to enable the “Hey Siri” feature on the latest iPhone and iPad Pro.

[[206926]]

Apple has integrated a very small speech recognition device into the motion coprocessor, which keeps running and waits for the user to say "Hey Siri". After recognizing these two words, Siri will process the subsequent speech as a command or query. The speech recognition detection device uses a deep neural network to convert the user's voice command into a rough distribution and uses time integration to calculate the confidence score of "Hey Siri".

If the score is high enough, Siri is activated and starts processing the user's command or query. If the score exceeds the maximum threshold, but does not reach the activation threshold. The device will enter "sensitive mode" for a few seconds so that when the user repeats "Hey Siri" again, it can be better activated. The addition of sensitive mode greatly improves the usability of the system and prevents strangers from accidentally opening "Hey Siri".

In order to reduce accidental activation of "Hey Siri" by default, Apple invites users to say these two words five times in a row when initializing "Hey Siri". During use, Apple will record the activation of "Hey Siri" in different environments, such as in the kitchen, in the car, in the bedroom, and on a tour. For more detailed information about "Hey Siri", please click here.