In the Fall 2016 semester, Sitara Persad, Andrew Xia, and Karan Kashyap worked on constructing models for the direct bi-directional classification of speech and images. For our final project, we trained two Convolutional Neural Networks to map image representations of digits to their spoken equivalent, achieving an image annotation accuracy of 88.5% and an image retrieval accuracy of 87.6%.
Our paper can be viewed here.