Pretrained language-vision models

CLIPSONIC is a novel approach to text-to-audio synthesis that leverages unlabeled videos and pretrained language-vision models. The study aims to address the challenge of acquiring high-quality text annotations for audio…