Google has been forced to publish a blog post elaborating on its speech technology practices after a few of its partners leaked confidential audio data to a Belgian news site. Product Manager David Monsees admitted that voice queries made to Google Assistant and similar products were being reviewed by itself and third parties, but only for the purpose of improving speech technology.
As part of our work to develop speech technology for more languages, we partner with language experts around the world who understand the nuances and accents of a specific language. These language experts review and transcribe a small set of queries to help us better understand those languages. This is a critical part of the process of building speech technology, and is necessary to creating products like the Google Assistant.
Belgian news site VRT received more than 1,000 leaked sound recordings and claimed it was able to identify people in the clips. The audio included everything from customer addresses to “a woman in distress and people talking about medical conditions.”