Belgian TV station VRT NWS has discovered that Google outsources listening to Assistant recordings to a third-party company, violating all privacy names
Beware: Google Home listens to you. One will say: what's wrong with that? Listening to the user's commands is the basis of how any self-respecting smart speaker works. The problem, however, is that the recordings of the audio picked up by the Google Assistant would have been sent to a third party company, in charge of transcribing the conversations in order to improve the speech recognition algorithm.
Google is spying on us?
VRT NWS came into possession of more than 1,000 audio clips recorded by Google Assistant via smart speakers and the Google Home app for smartphones. These audio files came with the users' addresses, so a VRT reporter got in his car and went to visit a bunch of Google users. He knocked on their doors, entered their homes, and played them the audio files. The users recognized their own voices and those of other people in their homes, including children. So much for anonymity and privacy.
Worse than Amazon?
This story is very reminiscent of what Bloomberg uncovered in April: Amazon employees replay recordings made by smart speakers and Alexa-compatible devices, once again to improve "speech recognition" algorithms. The difference between what we know today about Amazon and when it emerged about Google, however, is pretty big: Amazon doesn't outsource listening to recordings, while Google does. And it also gives them our home address. The third big player in the smart assistant market, Apple, also listens to our recordings again. But at the moment we don't have enough information to say how it does it.
Seeing the seriousness of the allegations, Google preferred to respond to VRT NWS: "We work with language experts around the world to improve speech technology by making transcriptions from a limited number of audio clips. This work is critical to developing the technology that makes products like the Google Assistant possible." According to Google just 0.2% of recordings are replayed and, the company explains the audio clips "are not linked to personally identifiable information." So how is it possible that the VRT NWS newsroom got hold of clips and home addresses? Google unloads the barrel: "We recently learned that one of these language experts may have violated our data security policy by leaking Dutch-language audio clips. We are actively investigating this and when we find a violation of our policy we will act quickly, up to and including closing our agreement with the partner."