#pretrained audio models

NVIDIA and MIT introduce Audio-SDS, a novel framework applying Score Distillation Sampling to audio diffusion models, enabling diverse audio tasks like synthesis and source separation without specialized datasets.