Hello, Crown Clients and Friends!
Often in the tech world, the more convenient something is then, the less secure or private it is. When processing power is available to automate or streamline everyday tasks, more information about our lives goes into helping the computer know (or guess) what we want. That information is the basis for targeted marketing, the theory that makes personally-identifying information into proprietary information. In this post, we’ll take a brief look at how voice recognition platforms, like Alexa, do their job and collect personal information so that you can make a more informed decision about how to protect your privacy.
Speech Recognition in Theory
For decades, speech recognition was considered too difficult and too processing-intensive to exist in consumer devices. The short version of the story is that speech recognition evolved alongside Machine Learning (ML) techniques over the past decade or so. Computers and software are now powerful enough to do a decent job at listening to the frequencies of our voice, the filter patterns of different sounds, etc., and quickly resolve these sounds into commands and words. Our cell phones, home automation systems, and desktops likely have speech recognition features that send speech data to the cloud for processing.
If we want to throw out a command to a computer with only our voice (and no button pushing), then that device needs to listen all of the time. Since Alexa (and other voice command software like Google Home’s Assistant, Samsung’s Bixby, etc.) is waiting to hear the “wake word,” which is “Alexa,” the software has to listen all of the time in case we say one. Alexa and Google Assistant routinely wake up and listen (that is, record us) because of erroneously hearing the wake word.
Where does this audio data go? It has to be processed for commands on tech companies’ servers, so it needs to be stored during processing. Since your Alexa device isn’t equipped to hear all of the possible words in the English language, and it doesn’t have enough storage to continuously record you, the audio data has to be shuffled off to Amazon’s servers. They then, by default, run checks with human workers to make sure the algorithm is correctly identifying words. This means that workers at Amazon are listening to what you say around Alexa.
Marketing versus Privacy
It’s possible that the convenience of turning on lights, checking traffic, or playing music with your voice is enough to put a little trust in the companies who make these devices and their software. It’s also likely that there’s not a lot to worry about here if the companies are routinely or immediately deleting (see below) the audio and only using it for voice functionality. It would be prudent, though, to consider what companies have been using this data for in the past.
In 2015, it was widely reported that TV commercials were being tagged with supersonic sounds, audible only to our cell phones, that could identify us as watching a particular broadcast. This information helps advertisers tag which shows or commercials you watch and add it to a profile of your consumer behavior that they can sell to other companies.
For years, users have reported that Facebook ads changed based on things that they had conversations about. One common anecdote that you may have heard: users sometimes left their phone next to a tv and turned on a foreign language broadcast, only to see their Facebook ads soon change into that language. For their part, Facebook has always denied listening in on conversations, but the proof has been available to show that someone was turning on the microphone to gather marketing information.
Having an appliance like an Amazon Echo or a Google Home invites companies to have an open microphone to listen, in your house, at any time. This is not to say that every company does listen to your intimate conversations, but they certainly could. A future change to the Terms of Service could make it so that you agree to have that information used for whatever the device manufacturer would like.
Restricting Alexa’s Audio Recording
If this kind of data collection worries you, you probably shouldn’t use a voice assistant at all, although that might not be feasible on some mobile operating systems (unless you’re interested in blocking the microphone). User-switchable restrictions on how this data is used do exist on some of these devices, which can limit the amount of data that is saved in the cloud. If you have Alexa devices, a good step is to go into the Alexa app, under Alexa Privacy Settings, How Long to Save Recordings, and tell it not to save recordings. Amazon will then delete the saved audio as soon as it is processed instead of having human workers listen to it, but again, we’re placing our trust in them to do so. For other voice-enabled devices, check the manufacturer’s support pages to see what you can do to limit data collection if anything.
-Written by Derek Jeppsen on Behalf of Sean Goss and Crown Computers Team