With the help of Voice-controlled Digital Assistants (VCDAs), end users can perform various tasks, such as creating shopping lists, setting reminders, or controlling smart home devices via voice commands. However, in multi-user environments, the different end users of VCDAs may not have access to the same controls to protect their privacy. The primary end users who set up VCDAs usually have full control over the data collected by VCDAs, including text transcripts and audio recordings of the other end users. In order for these secondary end users to gain access to privacy settings, they must also create an account with the appropriate manufacturer and accept an invitation from the primary end user to join the respective VCDA. As a result, they depend on the primary end user and the creation of a user account to be able to protect their privacy. Through a user account, however, personal information, such as name, address, or age can be linked to audio recordings, that poses additional privacy risks to secondary end users. For both primary and secondary end users, audio recordings are still maintained on cloud servers operated by manufacturers, resulting in a lack of transparency for all end users. In this paper, we thus propose an approach to improve the protection of both primary and secondary end users that reaches from the device set-up to its utilization. Our approach is based on the concept of a local registration and offline storage of voice commands.

mar

Pisa, Italy

https://ieeexplore.ieee.org/document/9767270/