Meta launches SAM Audio, an AI model for intuitive sound segmentation and isolation

Meta has launched SAM Audio, a state-of-the-art artificial intelligence model that brings advanced sound segmentation to users. This tool enables separating specific sounds from complex audio mixtures, such as isolating an instrument or voice within a song. Users can employ intuitive text, visual, or time-based prompts to filter out unwanted sounds, like background traffic in a video or a barking dog in a podcast.

Following the trend toward simplifying audio editing, SAM Audio reflects how people naturally interact with sound. It aims to make professional-grade audio separation much more accessible for creators, hobbyists, and non-experts. Previously, audio editing relied on a fragmented ecosystem of single-purpose tools, but SAM Audio unifies these features under one model and delivers advanced results across diverse, real-world audio scenarios.

Beyond instrument and vocal isolation, core use cases include noise filtering and supporting a broad range of creative audio workflows. For those interested in exploring its capabilities, SAM Audio is now available for download or can be tried in the Segment Anything Playground.