Study: Smart Speakers Are All Ears

One in every four adults in America now owns a voice-activated smart speaker.

USAToday reports Researchers at Northeastern University and the Imperial College of London spent the last six months streaming 125 hours of popular Netflix TV shows to a handful of voice-activated smart speakers. Their goal, which you can read about on a just-published webpage, “When Speakers Are All Ears,” is to figure out what words accidentally activate smart voice assistants – from Apple, Amazon, Google, and Microsoft – the most and what implications it might have on our privacy.

For the study, researchers built a custom cubicle and tested five types of speakers: a first-generation Google Home Mini, a first-generation Apple HomePod, Microsoft’s Harman Kardon Invoke, and two Amazon Echo Dots – a second-generation and a third-generation model. For the experiment, they binge-played TV shows including "Gilmore Girls," "The Office," "Dear White People," and "Narcos." They used a camera to detect when the speakers lit up, a microphone to monitor what audio the speakers played, such as responses to commands, and a wireless access point to record "all network traffic between the devices and the Internet."

The Amazon Echo dot device in the center is lighting up, signaling a voice activation

“We want to find out what exactly these systems are recording and where it’s going,” David Choffnes, an assistant professor involved in the study at Northeastern, told USAToday over the phone. “People fear that these devices are constantly listening and recording you. They’re not. But they do wake up and record you at times when they shouldn’t.”

In short, you’re not crazy. The smart speakers accidentally activated as many as 19 times a day and stayed awake – potentially recording and/or exposing private conversations unbeknownst to users. Around half of the accidental activations lasted less than six seconds.

The bad news: The other half recorded for as long as 43 seconds of audio each time. “We have found several cases of long activations. Echo Dot 2nd Generation and Harmon Kardon Invoke devices have the longest activations (20-43 seconds). For the Homepod and the majority of Echo devices, more than half of the activations last 6 seconds or more.”

Choffnes is quick to point out they found no evidence that these accidental recordings are being used in any nefarious way. “We haven’t found this to be a major privacy issue,” he said. “But we have a lot more work to do. We want to know how many activations lead to audio recordings being sent to the cloud vs. processed only on the smart speaker and whether cloud providers correctly show all cases of audio recording to users.”