Voice-powered speakers like Amazon Echo and Google Home have carved out a place on kitchen counters and nightstands in countless homes. What makes their immense popularity all the more remarkable is that they’ve achieved it without a key feature: Knowing exactly who’s talking. That changes with Google Home’s introduction of support for multiple accounts.
Starting today, your Google Home device will be able to identify up to six distinct voices.
Starting today, your Google Home device can identify as many as six voices, and summon information based on each person’s calendars, services, and preferences. In doing so, Google’s speaker-bound personal assistant becomes truly personal, unlocking the true potential of the most promising new category of consumer tech to come along in years. Oh, and it gives Google an important edge in the race with Amazon’s Alexa to own the next generation of interfaces. For now, anyway.
Look Who’s Talking
The benefits of a Google Home that knows you from your roommate, or significant other, borders on self-evident. The inability to match voices with identities means being unable to recognize the variety of preferences that fill any household. “There are so many services, if you think about the kinds of things an assistant should be able to help you with, that you want to be able to tailor to your experiences,” says Gummi Hafsteinsson, who leads the Google Assistant effort.
In practice, this means that once you’ve linked your account, saying “OK, Google, good morning” will conjure details from your calendar and news briefing preference, rather than defaulting to those of whomever set up the device. Extrapolate a bit further, and each individual in a household can summon everything from playlists to lighting cues to commute times tied to their specific routines. These fine-tuned interactions benefit Google, too.
“If Alexa or Google Home is going to develop a conversational relationship with you, it needs to be able to know who you are, which things you’ve asked for in the past, what other content and services you have access to,” says Forrester Research analyst James McQuivey. “In short, the full potential of intelligent agents can’t be realized without individual voice recognition.”
In fairness, you can change user accounts on Alexa-powered devices by going into the app. (And there’s little question Amazon also is working on the sort of feature Google just announced.) But the added friction obviates much of what makes digital voice assistants so pleasant to begin with.
It’s such an obvious win, in fact, it raises an obvious question: What took so long?
It’s not that computers can’t already tell one voice from another. They just haven’t had much reason to. “Until relatively recently, it was mostly irrelevant in terms of how the technology works,” says Brian Langner, a speech scientist at computer conversation company Pullstring. “It was usually a single microphone that tried to pick up all the various things it could, identify the speech, and figure out what the words are.”
Alexa and Google Home Record What You Say. But What Happens to That Data?
Amazon Alexa Hits 10,000 Skills. Here Comes the Hard Part
How to Keep Amazon Echo and Google Home From Responding to Your TV
Siri’s Not Even the Best iPhone Assistant Anymore
Samsung’s Bixby Hopes the World Needs Another Voice Assistant
Think of it in terms of dictation software. You speak, it listens and regurgitates. You don’t often see dictation by committee. Before today’s update, Google Home and Alexa applied that logic to far more complex interactions, from any number of speakers in a household. The devices continued improving their ability to know what you say, but haven’t cared much who was saying it. At most, you would expect a computer to know a single voice, and block out the rest.
That served Google perfectly well on Android phones, where voice recognition allows you and you alone to unlock your device simply by asking nicely. Home presents a significantly different set of problems. “The nice thing about a phone is that it’s reasonable to assume that there’s only one person using it,” says Alex Rudnicky, a speech recognition expert at Carnegie Mellon University. “If you have something like an Echo, it’s not entirely clear who the user population is.”
Ultimately, thanks to an assist from new neural networks running on the device, Hafsteinsson’s team smartened Google Home to the point that it just needs you to say “OK, Google” and “Hey, Google” twice each. With that, it knows you. Your voice also never leaves the device; Home does the identification onboard, and fetches the relevant info from Google’s servers.
“It wasn’t necessarily that we couldn’t do it,” Hafsteinsson says, explaining the gap between Home’s release and the introduction of multiple accounts. “It just takes time to get the quality level that we thought was good enough.” Such concerns also informed the number of users Home can identify. Six seemed big enough to accommodate most households, without straining the device’s capabilities. That makes sense, given the difficulty of discerning and identifying too many voices—a problem even for humans.
“When you start talking about constrained bandwidth situations, such as over the telephone, where you don’t have high-fidelity audio, even people are not necessarily great at identifying simply from the audio different voices that might sound somewhat similar,” says Langner.
What the Google Home still doesn’t do, though, is ignore voices it doesn’t recognize to begin with.
Private Little Chat
Burger King sparked a flame-broiled furor recently with an ad explicitly designed to set off Google Home devices. As the camera pans in, an actor says “OK Google, what is the Whopper burger?” Because Google Home takes its commands from anyone, it would spout off the first sentence of the “Whopper burger” Wikipedia entry in response, until Google shut it down.
This update wouldn’t have stopped the Burger King stunt. If another company tried it tomorrow, it would work. That’s by design; Google so far still prioritizes the ability to, say, allow dinner party guests tweak a playlist. “There’s a set of things that aren’t necessarily tied to a particular account that can still work,” says Hafsteinsson. “We’ve felt that there’s a ton of convenience about anybody being able to use those services.”
There still could be value in at least giving users the option to only let Home (or, eventually, Alexa) play nice with authorized voices.
“If I have my credit card registered with Amazon, there has to be some sort of guarantee that if stuff gets ordered that the proper charging is done,” says Rudnicky. Anyone with kids knows the danger of leaving them unattended with a voice-activated birthday list fulfillment machine.
There’s always the chance Google will include the ability to lock down your voice-assistant device in an update, or that Amazon will bring it to Alexa. For now, though, it’s not a big enough deal to turn you off to Google Home. Especially now that once you turn it on, your digital assistant will know just how to assist you.