Apple thinks of using artificial intelligence to improve "Hey Siri"

Apple has released a new edition of its machine learning magazine, where one of the successes of its digital assistant, Siri , is the ability to start listening as soon as the words "Hey Siri" are heard, as well as future enhancements to the feature.

The journal points out the progress of artificial intelligence in this area, which could eliminate Siri's need for initial training. Siri's team notes that some of the biggest challenges are unintended activation and activation. Problems by setting up Siri with the voice of the device owner.

Apple also requires users to train Siri briefly through five phrases creating a user profile stored by the device, and Siri then quietly adds 35 acceptable words used by the user to improve the profile, and interesting about the words "Hey Siri" Tries to match one sound against the subsequent repetition of the same phrase; instead, the profile contains a basic copy of the phrase with 40 differences calculated arithmetically.

Siri is active if "Sir Siri" is equivalent to or higher than the average score, and the device stores recordings of those 40 phrases for "Hey Siri", which allows re-creation of the user profile without the user retraining at any time, while working Apple improves the system by updating the software.

Apple expects in the future that Siri will not be trained in advance, so that the profile of the user will be empty at the beginning, and then use the user permission to obtain the required forms, updating itself by the emergence of additional requests, and also works on ways to detect false approvals , Which triggers the Siri activation by things other than the user.

The iPhone maker believes it can reduce the rate of fake approvals by almost 40 percent from 50 percent, the current rate when user says "Hey Siri" but the device does not respond, using deep neural networks, as well as reduce the possibility of activating Siri By another person by almost 75 percent.

Apple is trying to improve the performance of Siri audio plugin in large rooms and noisy environments and open air, and the company's team that he is currently looking at ways to deal with environments that are not closely aligned with Hey Siri recordings of the user profile, so successful in the so-called multi-style training, Increase the subset of training data with different types of noise and echo.

Throughout the year, Apple has been making changes to its voice assistant, not necessarily talking about all the adjustments, so it is not clear when these enhancements will be implemented. However, it highlights the major changes that have been made to Siri's Every annual developers' conference in June, where some key announcements are expected in the not too distant future.