Some smart speakers listen to conversations up to 19 times a day

Amazon, Google, Apple and Microsoft know a lot about you because of your internet searches, the data you share in their applications or devices, but also because they record your conversations with Alexa, Siri, Google Home and Cortana.

Each of these speakers has keywords with which you activate and interact with them. However, on many occasions, these activation words are not pronounced and the intelligent speakers are activated equally, without your permission and without notifying you. And they can do it up to 19 times a day according to a new study from Northeastern University, in Boston, United States.

Each of them responds to different random words - that is, not all are confused equally - and the average of activations per day is between 1.5 and 19 times, according to the study. Of these, the maximum recorded conversation has been 43 seconds and the minimum 6 seconds.

Siri and Cortana are activated more times, but the second-generation Echo Dot records more seconds

The research team at Northeastern University (Massachusetts, United States) has tested 7 smart speakers: Google Home Mini and Apple HomePod first-generation, Harman Kardon Invoke from Microsoft, and 2 pairs of Amazon Echo Dot second generation and Amazon Echo Dot third generation.

To do this, they used several of your favourite Netflix series with enough dialogues to expose smart speakers to a studio simulating everyday conversations. In total, 125 hours of dialogues of Grey's Anatomy, The Office, Dear White People, Big Bang Theory or Narcos, among others.

To detect when these are activated and begin to listen and record conversations, they used a camera to detect if they were illuminated when activated, network traffic to see if they sent data to the cloud and reported recordings of the speakers to the cloud if they were available.

Of these, the worst results were the Apple HomePod and Microsoft Cortana, followed by Echo Dot series 2, Google Home Mini and Echo Dot series 3 - although they do not specify the exact activation frequency for each of these found between 1.5 and 19 times in 24 hours.

However, although Siri was the one that was activated most times, he only used to listen to 6 seconds of conversation, as did the 3rd generation Echo Dot. But Cortana and the 2nd generation Echo Dot listened between 20 and 43 seconds. And, although some of the words that they confused can be quite obvious - that is, you might even confuse them - others are very rare and do not coincide at all.

For example, Siri was activated when some words began with "s" or had the syllable "ri" as "Hey, sorry" or "I see" - "Hey, I'm sorry" or "I see" in his translation into Spanish. And that is evident. But he also identified his words of activation when they said "Faith’s funeral", which had nothing to do with it. Or Cortana activating when they said, "Colorado" or Alexa with "Kevin’s car" and "congresswoman" ("Kevin's car" and "congressman").

The study was conducted with English-speaking programs, but that smart speakers are confused with any word other than your wake-up call, it can happen in any language. Of course, the study does not confirm that conversations are recorded whenever the speakers are activated, although a large part of them.