AI Organizations Release New Video-Audio Dataset

According to PANews, prominent AI institutions Grass, Ontocord, and LAION have jointly announced the release of the VALID (Video-Audio Large Interleaved Dataset) dataset. This dataset is constructed based on Grass's video repository and includes 30 million audio segments. These audio segments are interleaved with images and text, making it the first video-audio interleaved dataset in the industry. The release of VALID is expected to provide new data support for the training of multimodal AI models.