x.com/aiatmeta/…

Meta’s AI at Meta account announced SAM Audio, the latest open-source model in the Segment Anything family, capable of isolating and editing specific sounds from mixed audio using text prompts, visual cues from video, or time-based spans. This builds on prior SAM models for images, videos, and 3D by extending segmentation to audio, automating tasks like removing background noise or extracting instruments without manual editing.