Understanding and Developing Trustworthy Audio

Have you ever thought about how much audio data is really ‘out there’ on internet sites like YouTube, TikTok, and other social media platforms?

These days, audio AI technology has evolved to the point that realistic voices can be reconstructed from old audio recordings. There are numerous benevolent applications for this technology such as helping give a voice to people who have had throat cancer or other speaking-related disabilities. The Hollywood movie start Val Kilmer is a great example. He uses technology developed by a UK startup called Sonantic that generates a synthetic voice that sounds just like him.

However, this same technology can also be used to create audio deepfakes that impersonate a person and the way they talk, in order to spread misinformation or blackmail people. Further, there are not many international standards for preventing or detecting deepfakes from proliferating, which is a cause for concern for modern democracy.

This interdisciplinary project examines issues like these with the aim to characterise how the public views their own human rights in the context of audio AI. From this project, we will be able to inform AI regulators, policymakers, and standards organisations about what can or should be done to help people feel safe with audio AI technologies.