Speech as Personal Identifiable Information

Speech and Language Technology (SLT) has become an essential part of our digital society. The automatic processing of speech data (e.g., transcription of what-was-said or biometric recognition of who-spoke) is beyond the reach of end-users. Cambridge Analytica (2018) is giving us merely a pre-taster on the capabilities to subvert safety measures and infringing upon the privacy of end-users. On the other hand, the complexity of the challenge for the industry to protect privacy in providing voice services has recently been revealed when leading players had to admit that recordings are analysed by human operators (2019).

Recently, the International Speech Communication Association (ISCA) formed the Special Interest Group "Security and Privacy in Speech Communication" (SPSC). Intended as an interdisciplinary platform, the SIG fosters exchange between leading industrial and academic players with the goal to reach standards and procedures that protect the privacy of the individual in speech communication while providing sufficient means and incentives for industry to exploit future innovative services.

Aim. Through speech and language, humans interchange all sorts of information, also personally identifiable information (PII). We seek to bridge between usability, SLT, policy & governance, and cybersecurity to build a common understanding and identify overarching interests, challenges and pathways to solve these.

Tangible topics for initial inquiries are:

• Can voices be anonymised?
• Does semantic security hold across speech & language data life cycles?
• How to make responsible research and innovation feasible for policy & governance in speech & language technology?
• Transparent voice and speech interfaces
• Build a shared understanding of concepts and perspectives.

The goal is to collaboratively draft a roadmap for scientific inquiry and challenges ahead. These will be published open access online in post-workshop proceedings.