This thesis investigates two sequential studies toward real-time music synthesis directly from images via learned cross-modal embedding mappings and presents a unified deep-learning framework. In Study I, we explored a ...