Meta had already released its popular SAM models that helped isolate objects in images, and it’s now released a similar model for audio.
The company announced SAM Audio on Monday, describing it as a state-of-the-art AI model that can segment and isolate specific sounds from complex audio mixtures. The technology allows users to extract particular audio elements—like a guitar from a band recording or vocals from a song—using simple prompts, potentially streamlining workflows for audio and video professionals across multiple industries.
Breaking New Ground in Audio Separation
SAM Audio distinguishes itself as the first unified model to support three distinct prompting methods that align with how people naturally think about and interact with sound. Users can employ text prompts by typing descriptions like “dog barking” or “singing voice” to extract specific sounds. The model also supports visual prompting, where users can click on a person or object in a video to isolate their corresponding audio. Perhaps most notably, SAM Audio introduces span prompting—an industry first that enables users to mark specific time segments where target audio occurs.
These prompting methods can be used independently or combined, offering flexible control over audio separation. Meta positions this as a significant departure from the fragmented landscape of existing audio tools, which typically serve single-purpose use cases rather than providing comprehensive functionality.
Broad Applications Across Industries
According to Meta, SAM Audio has potential applications spanning music production, podcasting, television, film, scientific research, and accessibility services. The company suggests use cases ranging from filtering traffic noise from outdoor video recordings to removing unwanted sounds from entire podcast episodes. Meta is already incorporating the technology into its development of next-generation creative media tools.
The model is now available through Meta’s Segment Anything Playground, a platform that allows users to experiment with the company’s latest models using provided assets or their own uploads. SAM Audio is also available for download, making it accessible to developers and researchers looking to integrate the technology into their own applications.