J. Hershey, H. Attias, N. Jojic, and T. Kristjansson
Perceiving sounds in a noisy environment is a challenging problem. Visual lip-reading can provide relevant information but is also challenging because lips are moving and a tracker must deal with a variety of conditions. Typically audio-visual systems have been assembled from individually engineered modules. We propose to fuse audio and video in a probabilistic generative model that implements cross-model self-supervised learning, enabling adaptation to audio-visual data. The video model features a Gaussian mixture model embedded in a linear subspace of a sprite which translates in the video. The system can learn to detect and enhance speech in noise given only a short (30 second) sequence of audio-visual data. We show some results for speech detection and enhancement, and discuss extensions to the model that are under investigation.
|Published in||Proc. of the Int. Conf. on Acoustics, Speech, and Signal Processing|