Google announced a slew of improvements and updates to Google Assistant at its I/O 2018 conference on Tuesday (May 8). Chief among them was a new celebrity voice, but soon Google’s digital companion won’t only sound more natural, but be more helpful, too.
A more natural voice
Google is using a technology called WaveNet to smooth out the Assistant’s voice for a more natural, conversational experience. In doing so, the company was also able to create six new voices — one of which just so happens to be that of singer/songwriter John Legend, which was heard in a demo video answering a question directed to the Google Assistant about the weather outside. The new voices will be arriving later this year, and Google says their creation is a testament to the progress the company has made on the way to representing many more accents and dialects in the future.
Google Assistant will soon be easier to talk to as well, thanks to two new capabilities: Continued Conversations and Multiple Actions.
Continued Conversations allows Google Assistant to respond to a succession of queries without the user having to say a wake phrase like “Hey, Google” before each individual one. This isn’t just Google Assistant indiscriminately listening to everything you say; even with Continued Conversations, the Assistant is still able to tell the difference between the things you’re saying to it and a separate chat you could be having with a friend.
It’s part of an increased emphasis on contextual cues in listening and responding. With Multiple Actions, Google Assistant will also be able to follow through on two or more commands listed in the same query. As Google was quick to point out, it’s not as simple as it might sound — the Assistant has to be able to understand syntax to properly answer questions, and often times this is trickier than simply separating two phrases on either side of the word “and.”
Continued Conversations is arriving for users in the next few weeks, while Multiple Actions is rolling out now.
MORE: Google I/O: What's New with Google Assistant, Android P and More
A more immersive visual experience
Google is revitalizing the Assistant on phone screens for richer experiences. In a live demo on the I/O stage, the Google Assistant responded to a query about an actress with a full-screen preview of their Wikipedia entry. Another request to lower the temperature on the thermostat in the user’s home provided a custom circular slider so the user could make fine adjustments after the command was carried out.
Google is also giving the Assistant a home base of sorts with a new page that can be accessed with a simple swipe up. This screen surfaces various relevant bits of information pertaining to the user’s day, like the weather, calendar appointments and so on. In that way it fits the purpose Google Now used to serve on the Android home screen, except a bit prettier and now found within the Assistant.
The new visual experience is making its way to Google Assistant on Android first this summer, followed by iOS later in the year.
Assistant will one day make calls for you
It’s not ready for primetime quite yet, but one of Google’s biggest plans for the Assistant actually involves making phone calls on your behalf. Google calls the project Duplex, and with it, you can tell the Assistant to contact a restaurant, for example, and make a reservation for you over the phone.
This isn’t just the Assistant leaving a message — in two pre-recorded stage demos, it actually carried on frighteningly lifelike conversations with the people on the other end of the line, dynamically responding to unexpected questions on the fly to settle appointments in a natural and conversational tone. In neither example did there seem to be any indication the employees had any idea they were taking calls from a computer.
Google CEO Sundar Pichai says the company is working hard to make sure it gets Duplex right before it launches, so it’ll be quite a while before the feature debuts in full. According to a deep dive on Duplex published on Google's AI blog, in testing the system is handing off to human operators when conversations become too complex for it to handle on its own — though Google says it is able to conduct the majority of its tasks "fully autonomously" in its current state.
