Smart Assistant Smarter
TOP TIPS FOR IMPROVING SIRI, GOOGLE AND ALEXA
"Yes Dave, I can do that! "
Every smart assistant, like every computer platform, is different in its own way. A Mac is not the same as a Windows PC or Linux box; although they share a vast amount in terms of concept and design, while they might use some of the same components or even (technically) some of the same underlying code, they are distinct.
They’re powered by very different engines, they have different levels of knowledge available to them, they have different hardware makeup and methods of connectivity. But one thing does tie Siri, Alexa, Google Assistant and minor assistants like Bixby and Cortana together: they can all be as dumb as a box of frogs when they feel like it.
Say what?
Whichever platform you’ve staked your claim on, there are things you can do to improve both your AI assistant’s ability to listen and its ability to deal with what it thinks it has heard.
If you use discrete hardware to talk to them – an Amazon Echo, Apple HomePod or Nest Home device, for example – you should be careful about where that device is positioned in the room. These tend to use an array of far-field microphones to listen for their wake word, but if they’re not catching the sound correctly they may well miss (or at least mishear) what you have been trying to say.
Begin by shifting the unit away from the wall slightly, so that they’re not catching too many echoes from nearby walls; this can also fix any network connectivity issues you might be having. Make sure there’s no noise in the vicinity that could drown out the sound heard by those microphones; fans, open windows, air conditioners and the like can block an assistant’s ears significantly.
If you’re using an Amazon Echo device, there’s a handy test you can do to see if any of the microphones might be blocked or obscured somehow – just call out the wake word from different locations in your room. The blue ring will light up, with one of its segments a brighter blue to indicate the direction it thinks the sound came from. A broken or blocked mic (and sometimes wall echoes) will cause issues with triangulation, and the indicator will point in the wrong direction. Try a little compressed air to clean out any blockages, and make sure there’s nothing obscuring the top of the device.
Third-party assistants support systems like Amazon’s Alexa and Google Assistant without necessarily using the same hardware, and thus may have even less acute hearing; be extra careful where these are placed. It goes without saying that, even with a microphone array listening all around, smart assistants can be baffled or misactivated by TVs and speakers. There’s some delicious irony in there, but we’ll just say that you’re better off leaving the volume at a lower level (or moving the hardware away from the source of noise) if you’re likely to be talking to your smart speaker a lot.
On a phone or PC/Mac, you don’t quite have the same positional luxury. Phones usually aren’t so difficult to deal with, because you can (and naturally will) pick them up and put them near your mouth; the lack of a far-field array is probably a benefit here, because it cuts down on most accidental actuations. In terms of Siri or Cortana on a compatible Mac or PC, voice actuation offers a little conundrum; do you fight to make sure you can always be heard, or ignore it altogether? Siri and Cortana can be brought up in a couple of clicks, or a tap of the MacBook Touch Bar. Adding an external mic like that from your headphones can help – make sure it’s selected in Sound Preferences – but is it really worth worrying about given that you’ll likely be physically next to your PC or Mac at any time you’d want to use it?
Testing, testing
Assistant stupidity is not always entirely the fault of microphones. The things you say to an AI are not processed locally by the device, generally – they’re sent off, worked on by some faceless computer in the cloud, and the results are sent back. This can be a problem. While AI agents have a
There are things you can do to improve your AI assistant’s ability to listen to commands.
vast (and ever-growing) supply of data to compare captured snippets of speech against, doing so quickly demands a more refined approach: they’ll choose a specific section of the available data.
Broadly, this is regional. In Australia your assistant will use the Australian dataset; if you’re in India, you’re likely to have a vastly different accent, so the assistant will work with a localised dataset – except when it doesn’t. Make sure your localisation settings are correct if you want to be properly understood; while it might be fun to have Google Assistant chat to you in an American accent, it might also make interpreting your own voice more difficult. On the flip side, both Alexa and Google Assistant now support multilingual mode. Alexa can understand (and talk back) in English and French if your region is set to Canada, for example. Google gives you a choice of any two of its supported languages.
You can also, on most assistants, perform some level of voice training. This doesn’t take long, but it further narrows the dataset that assistant looks at when working out what you’ve said, and allows the assistant to tell you apart from other people that might use it. That’s great, except when it isn’t: there’s some small anonymity when talking to an untrained assistant, barring the fact that whatever you say is linked to the logged-in account. You can probably skip voice training if you’d rather your assistant didn’t get all HAL 9000 and start calling you by name.
Speaking of names, you can change Alexa’s wake word if you’re getting a few too many accidental actuations; there’s the delightfully Star Trek-esque ‘Computer’, as well as ‘Amazon’ or ‘Echo’ on offer either by asking the assistant verbally or by changing a setting in the app. Amazon seems to be the only company to have built this function into the limited circuit used to listen for a wake word, so if you’re using Siri or Google you’ll be stuck with the standard names.
Each assistant has its own level of extendability. Siri, for example,
You can probably skip voice training if you’d rather your assistant didn’t get all HAL 9000 and start calling you by name.
has its particular set of commands and abilities, but not a huge amount of third-party extras – while the likes of Alexa can be added to pretty easily.
That only paints part of the picture, though. Yes, you can add smarts to Alexa through its vast array of so-called skills, but they are of varying usefulness. Sometimes they’re required, like when you add a new piece of hardware to your home. Sometimes they’re just fun little excursions. But very rarely are they useful. It’s possible you’ll find something for your niche interests, but the quality of third-party additions varies.
Siri’s close integration with the Apple ecosystem actually makes it a little smarter with a little less effort. Add devices to your HomeKit network and it’ll know about them. With only a little fiddling, you can teach it the relationship of you and your contacts.
Many consider it to be a little more inflexible than its rivals, but that really depends on how you’re going to use it. With their cheap devices, Google and Amazon make it much easier to fill your home with smart assistance, but again, Siri’s know-it-all attitude to Apple devices (and deep integration) makes it a much smarter cookie if you’re all-in on Apple.
Smart follow-ups
Want to really make your smart assistant smart? Smarten yourself up. Recent developments means all three major players have a whole lot more synapses firing, with quality of life improvements many people don’t even know are there.
Both Google Assistant and Alexa now support some kind of follow-up mode. With Alexa you’ll need to switch this on in the app; for Google devices, it should be enabled by default. It means you can combine commands – “Hey Google, play Van Halen and set the volume to 10” – or take advantage of a brief few-second listening period after you’ve asked a question. If you ask how the weather is and get a typically negative answer, you can set a reminder to bring an umbrella without needing to use the wake word again.
Even if you don’t use something like follow-up mode, learning how best to talk to an AI assistant makes the whole process feel more slick.
Contrast the laborious process you’ll have to go through after you say “Alexa, set a reminder” (and it is a long-winded one, at best) with simply saying “Alexa, remind me to go for a walk at 6pm tomorrow”.