Meta has unveiled SAM Audio, an advanced AI model designed for sound segmentation, enabling users to isolate or remove specific audio elements from complex sound environments. This product is aimed at content creators, musicians, podcasters, video editors, researchers, and anyone requiring precise audio manipulation. SAM Audio is now accessible via the Segment Anything Playground platform and can be downloaded for broader use, making it available to both technical and non-technical users worldwide.
🔉 Introducing SAM Audio, the first unified model that isolates any sound from complex audio mixtures using text, visual, or span prompts.
— AI at Meta (@AIatMeta) December 16, 2025
We’re sharing SAM Audio with the community, along with a perception encoder model, benchmarks and research papers, to empower others to… pic.twitter.com/FuMJyULmJR
SAM Audio supports three unique prompt methods:
- Text Prompting: Users describe the sound they want to isolate or remove.
- Visual Prompting: Allows clicking on objects or people in video frames to extract their associated audio.
- Span Prompting: A new approach that lets users specify time segments to target particular sounds.
This combination provides users with granular control over audio separation, setting it apart from previous tools that typically focused on single-use scenarios or required technical expertise.
Compared to earlier audio editing solutions, SAM Audio stands out for its integration of multimodal prompts and its real-time, user-friendly workflow. Early testers have noted the model's high accuracy in separating overlapping sounds, and industry analysts highlight its potential impact on accessibility and creative media production.
Meta developed SAM Audio as an extension of its Segment Anything initiative, which aims to democratize advanced media editing tools through open access and AI-driven models. The release reflects Meta's commitment to supporting a global community of creators and developers by offering cutting-edge AI resources for diverse applications.
