Real-World Sound Dataset Project Secures Funding to Help Make AI Safer
Dr Iran Roman, Lecturer in Sound and Music Computing at the School of Electronic Engineering and Computer Science, has been awarded up to £100,000 by the AI Hub in Generative Models to develop a dataset that could help make AI systems safer, more reliable and better aligned with the real world.
/filters:format(webp)/prod01/channel_164/school-of-electronic-engineering-and-computer-science/media/eecs/ai-hub-in-generatie-models.jpg)
A member of the Centre for Multimodal AI and the Centre for Digital Music, Dr Roman's research sits at the intersection of neuroscience, machine perception and AI, exploring how machines can better understand sound in the way humans do.
The project will create a dataset capturing how sound behaves in everyday environments - such as the home - combining 3D audio, video and precise 3D object tracking. Most AI systems today are trained on simplified or synthetic data, limiting their ability to interpret complex, noisy, real-world situations. This dataset aims to change that, with applications ranging from detecting household hazards to helping robots navigate safely around people and pets. The work is supported by industrial partner Meta.
The project is one of four to receive a share of £400,000 through the Hub's Dataset Creation & Challenge Projects programme. David Barber, Director of the AI Hub in Generative Models, said:
"This project seeks to address a genuine gap. We are excited about the possibilities that the dataset created through this project would provide and the potential benefits it could achieved, with a little help from the tech community."
The AI Hub in Generative Models is one of nine UK AI research hubs funded by EPSRC as part of UKRI's AI programme.
You can find out more on the AI Hub in Generative Models website.