Event Details

Get fresh perspectives and think about new ways of approaching your own research, meet new people and explore potential research collaborations. Come be part of the DSAIL community!

How does Integrity sound?

In tune with DSAIL’s new Integrity theme, this Data Dialogues session discusses recent and upcoming projects that specifically confront the integrity of voice data. The dominance of generative AI has prompted us to question the integrity of voice data in calls from the bank, in election campaign videos online, and in audio recordings used to incriminate someone. While the prospect of using generative AI to mass-produce voice data could benefit a range of industries and sectors, it is easy to see how these AI-generated speech samples could be deployed for malicious purposes. Researchers cottoned on to this possibility long before we reached the qualities of AI-generated speech samples we hear today, and so there is a substantial body of ongoing risk-focused research that seeks to mitigate the risks and harms of convincing AI-generated speech samples.

This session exchanges presentation slides for audio recordings to informally walk through the recent developments in synthetic voice datato demonstrate how “risky” these data have become. At the same time, this session introduces some newer ideas for lesser-known opportunities that could come with AI-generated speech data (such as AI-generated speech synthesis systems as Privacy Enhancing Technologies (PET), as well as explainability tools in forensic voice identification).

In light of the above, this session hopes to tempt attendees to participate in Lancaster University’s HackaCon – a hackathon that steers risk-focused research on AI-generated voice data in a new direction.