Audio-Visual Scene Recognition

Traditionally audio and visual recognition departments have operated independently, however with the emergence of new deep learning techniques there is a possibility to combine elements from both sets of research groups, in the audio and visual domain.
My proposal is to work on a novel technique that will allow a system to draw on image and speech recognition in order to describe a scene by exploring the recent research in Deep Learning, specifically, Cycle GANs. There are many potentially useful applications for such a system, such as YouTube videos or for companies such as IKEA for video instructions on constructing furniture.