On-device speech recognition on smart speakers: Only half of the cloudless smart home equation

When Google announced its “end-to-end, all-neural, on-device speech recognizer” for the Gboard phone keyboard two weeks ago, I was initially excited. That’s because speech recognition can now run locally on Gboard, even without a network connection.

So my next thought was that it’s not really a stretch to move that same neural network functionality to Google Home products. This would allow for the Google Assistant to recognize voice input even if your home internet connection is down. Sure, in this case, the Assistant can’t get external data from the web such as the weather or traffic information, but that’s OK: It could still accept smart home voice commands.

And then it hit me: Even if a Google Home device could interpret my saying “Set the thermostat to 70-degrees”, in most cases it really can’t comply. That’s a shame because we’d be much closer to a cloudless smart home if it could.

Many smart home owners don’t want to be completely reliant upon the cloud for various reasons. It could be that they don’t want all of that data moving beyond the confines of their home and have it either intercepted by a bad actor or sold off for some business reason. Others simply want total control of their smart home and often go with a local hub or server solution. And some don’t want to deal with the occasional outage associated with a particular smart home product company.

So again, I was hopeful that edge-based natural language processing would be of help here. Alas, we’re still stuck with the cloud for the time being; at least those of us who built a smart home around smart speakers instead of a purpose-built hub such as a Hubitat Elevation, a Samsung SmartThings or a Wink hub.

To test this, I simulated an “offline” situation by removing the ethernet cable running from my FiOS router to my Google Wi-Fi device. I still had a Wi-Fi connection for my devices and computers to talk to each other but they couldn’t speak to the cloud. Since I was simulating local controls, I went into my Google Home app and tried to turn on some lights and adjust the thermostat. I had to use the app because language processing on the Google Home products still relies on the cloud, of course.

And the result? No cloud, no device control, of course, unless I used my Wink app; the Wink hub supports local automations and control.

With both of the main smart speakers – Amazon Echo and Google Home products – relying largely on connected cloud accounts, this situation isn’t going to change any time soon. What needs to happen is fewer connected service accounts, not more. And yet, each time I go into my Google Home app, I seem to see scores of additional product options that link to the smart speaker.

I understand why we see more of these: Google wants to bring the Assistant service to the widest number of smart home products as possible. So too does Amazon, for that matter. And these are both big players in the cloud services business with Amazon’s AWS leading the profitable pack. Both have been moving additional services to the cloud for the last several years.

At some point, however, I suspect (or hope, really) that at least one of the two will move from a “let’s link an account with every company you have a smart device for” to some standard IoT platform that product partners can integrate with and bring more control to the local level.

We’re reaching the point where we don’t need much more device support in the smart home; it’s time to address consumer concerns and bring more localized control to the smart home, even if we use a smart speaker.