Digital voice assistants: A new kind of user agent

Digital voice assistants provide a new way of browsing and interacting online. Now we can simply speak and listen to our devices, almost as if to a human assistant. However, the convenience of using our voice comes at a cost: On a cue, audio of our speech is streamed to the cloud of the service providers, where they are retained and processed at their discretion. These speech data contain more information about us than simple text, including features of our voice, which can be used to identify us, and of our mood and energy level, which can be used to profile us to an extent that may not be justified by the convenience we get in return. To mediate between the interests of the user and those of the service providers, the General Data Protection Regulation (GDPR) puts the user in control of their personal data. In this paper, we show that speech data qualify as personal data according to the GDPR; however, in practice service providers can operate around the regulation, and so, it appears ineffective. This is risky given how forcefully voice assistants are pushed to market by companies who, by many accounts, operate under surveillance capitalistic objectives to control user behavior.