Manufacturers are experimenting with a range of shapes, sizes, prices and features, in the hopes of connecting with consumers — from $400 boom boxes to alarm clock replacements. Having persuaded customers to allow constantly listening microphones in their homes, tech companies are now adding internet-connected cameras and screens to their line-ups, testing the limits of consumer tolerance around privacy and security.
The wide range of hardware becoming available shows, even as Silicon Valley seizes on artificial-intelligence powered speakers as the next big thing, the industry is yet to figure out exactly how people will use these devices — or whether they will keep buying them once the novelty of a talking kitchen assistant wears off.
“They are still throwing things at the wall to see what sticks,” says Carolina Milanesi, analyst at Creative Strategies.
Voice recognition and language understanding technology has “gone from being pretty bad to being far better in the last two years”, says Phil Libin, former chief executive of Evernote and currently an investor focused on digital assistants and AI. Now, the bigger problem is whether it can understand the user’s needs and provide a useful reply. “That’s really variable and depends on context.”
While the built-in functions of speakers such as Amazon Echo and Google Home work well, third-party services and integrations, which Amazon calls Alexa’s “skills”, are still “very clunky”, Mr Libin says. “That’s the bigger thing I worry about as an investor.”
In surveys this year, Creative Strategies found that more than half of early adopters of Amazon’s Echo speakers had not increased their usage of the devices, with 22 per cent using them less over time — even as the range of “skills” or apps has grown.
“That is a bit concerning, so early in the cycle,” Ms Milanesi says. “You would think that if you’re discovering what else you can do, you would therefore use them more.”
The survey also found that the single most popular use for Alexa was not shopping or controlling smart-home devices but something much more basic: setting an alarm. Products offering this feature are available for much less than the Echo’s original $180 cost.
Ms Milanesi says this reflects users’ hesitancy to experiment with a product that can still be unreliable in its responses.
“We use the timer all the time in the house because it’s easy, it works and [Alexa] gets it right every time,” she says. “That is what I think consumers are still looking for.”
While analysts believe that Google’s Home speaker is trailing far behind Amazon, Google is drawing on its long experience in search and personalisation to improve its Assistant.
“We’ve trained it as only Google could, with 50m voice samples from hundreds of different ambient environments,” said Rishi Chandra, general manager of Home, as he introduced new products this week. Its ability to “voice match” individual members of the household as they talk to its speakers is “a really big deal”, he said. “An assistant can only be truly useful if it knows who you are.”
Another area ripe for development is bringing truly conversational dialogue to virtual assistants, allowing a more realistic back-and-forth between human and machine. “In the consumer market for virtual assistants, there isn’t much of a dialogue, it’s more transactional,” says Robert Weideman, executive vice-president at Nuance, a pioneer in natural-language processing.
That presents a challenge that goes beyond raw technology and into the fuzzier art of user experience design.
“I don’t think most people know how to design for conversations yet,” says Mr Libin. “It’s the type of user experience we don’t have a long history with.”
Despite Google’s repeated pledge to rethink all its products for an “AI-first world”, Mr Weideman suggests that Amazon’s data about behaviour from online shopping would ultimately prove an advantage.
“[Google’s] DNA is different from Amazon, whose business model is engagement for purchasing, so their data are much more robust,” he said.
Many of the latest hardware updates involve something rather more straightforward: improving sound quality. Manufacturers are doubling down on playing music, which is one of the most popular uses for AI-connected speakers.
After Apple emphasised the high-fidelity sound of its forthcoming $350 HomePod, Google responded this week with Home Max, a supercharged version of its Assistant-powered speakers costing $400. Amazon has also upgraded to Dolby-powered speakers in its new $150 Echo Plus.
Sonos, a pioneer in connected speakers, also introduced its first voice-controlled devices this week. Sonos saw its growth rate take a hit from the rise of smart speakers, but now it is racing to embrace the technology.
As well as adding the new $200 Sonos One, which will support both Alexa and Google Assistant, its existing speakers can now also be controlled using an Amazon Echo.
“We have heard loud and clear from customers that they want to use voice to control their Sonos,” says Patrick Spence, who took over as Sonos chief executive in January. “Voice just makes it even easier to get the music playing.”