Meta has introduced SAM Audio, a new AI model that extends its Segment Anything framework from images to sound. The model allows users to separate specific audio elements—such as speech, music, or background noise—from audio or video using simple prompts. Built as a generative audio separation system using a diffusion transformer, it produces both target and residual audio streams. Meta says the model achieves state-of-the-art performance across prompting methods and is releasing an open-source