Machine Learning Journal: Apple Explains the Origin of “Hey Siri”

Apple’s Siri team has a new entry in the Machine Learning Journal released. The first entry for this year is again about Siri and how the voice activation of Apple’s AI Assistant works.

“Hey Siri”

After Apple last fall, part of the process documented As Apple’s voice assistant has responded to the call “Hey Siri”, the Siri team is now giving another look into the feature. This time, the developers explain, among other things, how they have implemented it, that Siri only hears the voice of the owner and how it came to the fact that the expression “Hey Siri” prevailed.

To prevent Siri being triggered unintentionally, the responsible development team focused on the accuracy of speech recognition. One thing that you really wanted to prevent is a stranger who says “Hey Siri”. To prevent unwanted activation, Siri should only listen to the owner of the iPhone.

For this, the user must first be made familiar with the system, which is done with the help of some “language exercises” during initialization. To recognize the user’s voice later, the development team worked on the voice analysis for a long time. Many tests have been done to detect false positive detection. Based on these results, the system could be adjusted and significantly improved.

But how did the phrase “Hey Siri” come about? As the developers explain, it was particularly important to the team that the voice activation sounds natural. The process should adapt to the normal spoken sentence so that the activation words do not feel like a foreign object in the actual task. To find the perfect “initiation” for a voice command, one quickly took a command that was already in circulation. So the developers observed that iPhone owners often said “Hey Siri” when they pressed the homebutton. Thus, the use of this “saying” was close.

Leave a Reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: