Google Assistant has historically relied on cloud-based services to recognize your voice and respond to your requests. But Google says it’s managed to shrink its voice recognition model to about 500MB, allowing it to be stored on a phone for on-device processing.
Since there’s no round-trip to the cloud required, that means Google Assistant can recognize and respond to your voice much more quickly.
In a live demo at Google I/O today, Google showed a Google Assistant responding nearly instantly to dozens of requests.
Among other things, that means opening apps, taking selfies, responding to incoming messages, composing emails, getting directions, or searching through photos.
Continued Conversation support means that you don’t need to say “OK Google” before every request. If you already opened Google Assistant to view your calendar, for example, you can then just ask it to add an reminder or start composing a message.
Some actions will also work even if you’re offline, such as toggling your flashlight or viewing your calendar.
Other announcements from Google I/O so far include:
- Google Duplex for the web
- New Google Lens features including highlighting popular items on restaurant menus, or calculating tips from receipts
- Camera and augmented reality features in the Search app
- Podcasts in Search