×

Ever wanted to casually chat with John Legend? Now you can, in the comfort of your own home no less, thanks to a new Google Assistant mode that makes use of Legend’s voice to respond to everyday queries about the weather, random factoids and more. However, there’s a twist to this: The responses aren’t just based on recordings of Legend. Instead, Google used artificial intelligence to recreate his voice for customized answers.

Google first announced that it would be bringing to Legend’s voice to the Assistant back in May of 2018. Almost a year later, the feature is finally available to U.S. consumers using the Google Assistant on their Google Home smart speakers, TVs or mobile phones.

Consumers can simply ask their smart speaker to “talk like a legend,” or change the voice of their Assistant in the relevant settings on their phone. After that, Legend will be able to give them weather reports, tell them jokes, and answer factoids about cooking and space.

Google also worked with Legend to add some extra easter eggs, which include lines about his wife Chrissy Teigen, spontaneous Happy Birthday serenades and more.

The search giant didn’t go full Legend for this collaboration; many everyday voice queries will still be answered by one of the Assistant’s regular voices. What’s more, Legend’s voice will also just be available for a limited time.

Still, even having Legend tell everyone in the U.S. the weather for the next couple of weeks sounds like a major undertaking, considering the myriad possible combinations of temperatures, precipitation and more.

That’s why Google relied on some of its A.I. smarts to teach the Assistant how to talk like Legend. Instead of having the singer record every single possible answer, the company spent some time in the studio with him to capture him reading various scripts, and then taught its Wavenet neural network the quirks of his voice.

Wavenet is the same technology that Google uses to make its regular Assistant voices sound more natural, and less robotic. In essence, Wavenet does so by analyzing existing speech for nuances like lip smacks, accents and more. It then applies that knowledge to any new speech it generates based on text input.

So when you ask your Google Home speaker to tell you the weather in a John Legend voice, it’s actually Google’s A.I. generating responses on the fly that sound like the singer. However, for queries like “sing me a song,” the response is still all of him.