Over the past three years, the personal assistant Google Assistant has been actively developed. Now available on over a billion devices, in 30 languages in 80 countries, with over 30 unique connected home devices from over 000 brands. The search giant, judging by the announcements made at the Google I / O developer conference, is striving to make the assistant the fastest and most convenient way to get results.
Google Assistant currently relies heavily on the cloud computing power of Google's data centers to power speech recognition and comprehension models. But the company has set itself the task of redesigning and simplifying these models so that they can run locally on a smartphone.
During Google I / O, the company announced the achievement of a new important milestone. Thanks to advances in the field of recurrent neural networks, Google has been able to develop completely new models of speech recognition and language understanding, as a result of which a 100-GB model in the cloud has been reduced to less than half a gigabyte. With these new models, the artificial intelligence at the heart of Assistant can now run locally on your phone. This breakthrough enabled Google to create a next-generation personal assistant that processes speech on the device with near-zero latency, in real time, even when there is no internet connection.
Running on the device, the Next Generation Assistant can process and understand user requests as they come in and provide responses 10x faster. This makes it much more efficient to complete tasks across applications, such as creating calendar invitations, searching and sharing photos with friends, or dictating emails. And with Continued Conversation mode, you can make multiple requests in a row without having to say “Ok Google” every time.
The next-gen assistant will be coming to new Pixel phones before the end of this year. Obviously, we are talking about the autumn Pixel 4, which will receive new chips with improved neuroplates that speed up calculations related to AI algorithms.