In this paper we present a creative approach to reconstruct 3D audio for multiple sources from a single channel input by detecting and tracking visual cues using supervised learning methods. We also discuss a similar approach for improving speaker’s classification from a video stream by employing both facial and speech likelihoods, or simply Multimodal Speaker Recognition on a video stream.
Videos assets are here: