SAM Audio
Walk through Meta's SAM Audio approach to audio separation: how text, visual and time-range prompts stack to pin down a target sound, and what the PE-AV encoder underneath contributes.
A unified audio segmentation and editing model from Meta that isolates and edits specific sounds inside complex mixes from a prompt. SAM Audio accepts three stackable prompt types: text, visual cues that mark the sounding object, and time ranges for when the sound occurs. Typical uses include separating instruments or vocals, cleaning outdoor recordings and removing podcast noise.
